Postanowiłem ulepszyć swoje zdolności w generowaniu filmów z modeli AI (lokalne, darmowe, open weights).
Przez długi czas korzystałem z dwóch kart NVIDIA Tesla P40 24 GB zainstalowanych w serwerze HPE ML350P gen8. Nadal świetnie sprawdzają się przy lokalnych modelach językowych (Dużo VRAM), jednak w przypadku nowoczesnych modeli do generowania wideo ich wiek zaczyna być odczuwalny. Brak rdzeni Tensor nowszych generacji oraz niższa wydajność obliczeń (FP16, FP8) sprawiają, że generowanie filmów trwa znacznie dłużej niż na współczesnych kartach GeForce. Dlatego zdecydowałem się wykorzystać leżącą odłogiem stację HP Z440 z 64 GB RAM i wyposażyć ją w GeForce RTX 5060 Ti 16 GB.
Wybór padł na ten model ponieważ ma relatywnie dużo VRAM w stosunku do ceny, jest to nowoczesna, relatywnie tania karta w architekturze Blackwell i idealnie nadaje się do budżetowej stacji AI. Podłączana jest do zasilacza przez adapter 2×6 pin PCIe -> 1×8 pin PCIe. Karta pełnowymiarowa z trzema wentylatorami, ma dobrą kulturę pracy, jest wyjątkowo cicha i maksymalny pobór mocy wynosi 180 W. W połączeniu z cichą stacją HP Z440, która w trybie jałowym zużywa tylko 50 W, taki system może działać jako serwer 24h/d.
Zmieniłem również workflow generowania filmów. Poprzednio używałem ComfyUI, bardzo elastyczny ale upierdliwy ponieważ trzeba tworzyć swoje pipeliny w formie grafów i szybkie testowanie jest uciążliwe. Teraz używam WanGP, który jest czymś w rodzaju AUTO1111 ale dla modeli video, prosta obsługa, GUI oparty na interfejsie Gradio. WanGP zoptymalizowany jest pod karty <24GB VRAM i używa wiele modeli nie tylko generujących video ale i obrazy i TTS.
WanGP został zainstalowany manualnie na środowisku anacondy
Bardzo ważne, flaga –listen pozwala łączyć się z przez przeglądarkę, nie tylko po localhost ale także z zewnątrz. Mogę dzięki temu łączyć się przez VPN z całego świata dzięki serwerowi Wireguard na Raspberry PI. Serwer jest sterowany przez terminal i klienta ssh. Uruchomienie i zatrzymanie usługi oraz inne polecenia:
Przeprowadzono testy zarówno w trybie Image-to-Video, jak i Video-to-Video. W przypadku transformacji Video-to-Video wykorzystano animacje fraktali wygenerowane w programie Mandelbulber jako materiał źródłowy. Dla modelu Wan VACE zastosowano prowadzenie za pomocą map głębi (Depth) oraz kombinacji Depth + Optical Flow, natomiast dla modelu LTX-2 wykorzystano prowadzenie wyłącznie z użyciem map głębi (Depth). Dodatkowo wykonano testy Image-to-Video na podstawie pojedynczego statycznego obrazu.
Transform this abstract animation into an ultra-realistic cinematic flight through the core of a spiral galaxy. Preserve the original camera movement and geometry exactly, but reinterpret every organic structure as gigantic clouds of glowing interstellar gas, dust filaments, stellar nurseries and gravitational distortions. The bright center becomes the blazing galactic nucleus surrounded by billions of stars. Add volumetric nebulae, cosmic dust, blue and golden emission clouds, gravitational lensing, subtle light scattering, realistic HDR lighting and immense scale. The camera accelerates smoothly toward the luminous galactic core. High detail, IMAX space documentary, NASA-inspired astrophotography, physically based rendering, cinematic color grading.
Parametr
Wartość
Model AI
VACE 1.3B
Model bazowy
Wan 2.1 Text-to-Video 1.3B
Typ generacji
Text-to-Video (LV)
Rozdzielczość generacji
832 × 480 px (480p)
Liczba klatek
81
Liczba kroków (Steps)
20
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
1.0
Sliding Window
81 klatek
Seed
716045761
Czas generacji
6 min 29 s
Transform this abstract animation into an ultra-realistic cinematic flight through the core of a spiral galaxy. Preserve the original camera movement and geometry exactly, but reinterpret every organic structure as gigantic clouds of glowing interstellar gas, dust filaments, stellar nurseries and gravitational distortions. The bright center becomes the blazing galactic nucleus surrounded by billions of stars. Add volumetric nebulae, cosmic dust, blue and golden emission clouds, gravitational lensing, subtle light scattering, realistic HDR lighting and immense scale. The camera accelerates smoothly toward the luminous galactic core. High detail, IMAX space documentary, NASA-inspired astrophotography, physically based rendering, cinematic color grading.
Parametr
Wartość
Model AI
VACE 1.3B
Model bazowy
Wan 2.1 Text-to-Video 1.3B
Typ generacji
Text-to-Video (DLV)
Rozdzielczość generacji
832 × 480 px (480p)
Liczba klatek
81
Liczba kroków (Steps)
20
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
0.85
Sliding Window
81 klatek
Seed
98966023
Czas generacji
8 min 37 s
Transform this abstract animation into an ultra-realistic cinematic flight through the core of a spiral galaxy. Preserve the original camera movement and geometry exactly, but reinterpret every organic structure as gigantic clouds of glowing interstellar gas, dust filaments, stellar nurseries and gravitational distortions. The bright center becomes the blazing galactic nucleus surrounded by billions of stars. Add volumetric nebulae, cosmic dust, blue and golden emission clouds, gravitational lensing, subtle light scattering, realistic HDR lighting and immense scale. The camera accelerates smoothly toward the luminous galactic core. High detail, IMAX space documentary, NASA-inspired astrophotography, physically based rendering, cinematic color grading.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (GGUF Q8_0)
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
121
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.5
Sliding Window
481 klatek
Seed
469932869
Czas generacji
2 min 59 s
Transform this abstract animation into an ultra-realistic cinematic flight through the core of a spiral galaxy. Preserve the original camera movement and geometry exactly, but reinterpret every organic structure as gigantic clouds of glowing interstellar gas, dust filaments, stellar nurseries and gravitational distortions. The bright center becomes the blazing galactic nucleus surrounded by billions of stars. Add volumetric nebulae, cosmic dust, blue and golden emission clouds, gravitational lensing, subtle light scattering, realistic HDR lighting and immense scale. The camera accelerates smoothly toward the luminous galactic core. High detail, IMAX space documentary, NASA-inspired astrophotography, physically based rendering, cinematic
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (GGUF Q8_0)
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
121
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.2
Sliding Window
481 klatek
Seed
698909751
Czas generacji
2 min 59 s
Transform this abstract animation into an ultra-realistic cinematic flight through the core of a spiral galaxy. Preserve the original camera movement and geometry exactly, but reinterpret every organic structure as gigantic clouds of glowing interstellar gas, dust filaments, stellar nurseries and gravitational distortions. The bright center becomes the blazing galactic nucleus surrounded by billions of stars. Add volumetric nebulae, cosmic dust, blue and golden emission clouds, gravitational lensing, subtle light scattering, realistic HDR lighting and immense scale. The camera accelerates smoothly toward the luminous galactic core. High detail, IMAX space documentary, NASA-inspired astrophotography, physically based rendering, cinematic
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (GGUF Q8_0)
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (EVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
161
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.8
Sliding Window
481 klatek
Seed
926616269
Czas generacji
2 min 45 s
Transform the animation into a breathtaking cinematic journey through the interior of an immense emission nebula. Preserve the original camera path and overall geometry, but reinterpret every fractal surface as colossal volumetric clouds of ionized hydrogen, oxygen and cosmic dust. Replace all solid structures with translucent gaseous formations, turbulent plasma filaments, glowing dust lanes, dense molecular clouds and illuminated stellar nurseries. The brightest regions become intense emission nebulae illuminated by nearby massive stars. Fill the scene with billions of tiny distant stars visible through transparent gas clouds. Use realistic blue, cyan, teal, magenta and golden emission colors based on real astrophotography. Create soft volumetric lighting, multiple layers of gas, subtle light scattering, realistic HDR exposure and immense astronomical scale. The camera performs a smooth cinematic flight through the nebula with gentle inertia and subtle parallax. The environment should resemble real NASA and ESA deep-space images, the Pillars of Creation, the Carina Nebula and the Omega Nebula rather than abstract fractals. Ultra-realistic astrophotography, IMAX documentary, physically based rendering, extremely high detail, temporal consistency, stable geometry, no flickering.
Transform the animation into a breathtaking cinematic journey through the interior of an immense emission nebula. Preserve the original camera path and overall geometry, but reinterpret every fractal surface as colossal volumetric clouds of ionized hydrogen, oxygen and cosmic dust. Replace all solid structures with translucent gaseous formations, turbulent plasma filaments, glowing dust lanes, dense molecular clouds and illuminated stellar nurseries. The brightest regions become intense emission nebulae illuminated by nearby massive stars. Fill the scene with billions of tiny distant stars visible through transparent gas clouds. Use realistic blue, cyan, teal, magenta and golden emission colors based on real astrophotography. Create soft volumetric lighting, multiple layers of gas, subtle light scattering, realistic HDR exposure and immense astronomical scale. The camera performs a smooth cinematic flight through the nebula with gentle inertia and subtle parallax. The environment should resemble real NASA and ESA deep-space images, the Pillars of Creation, the Carina Nebula and the Omega Nebula rather than abstract fractals. Ultra-realistic astrophotography, IMAX documentary, physically based rendering, extremely high detail, temporal consistency, stable geometry, no flickering.
Transform the animation into a breathtaking cinematic journey through the interior of an immense emission nebula. Preserve the original camera path and overall geometry, but reinterpret every fractal surface as colossal volumetric clouds of ionized hydrogen, oxygen and cosmic dust. Replace all solid structures with translucent gaseous formations, turbulent plasma filaments, glowing dust lanes, dense molecular clouds and illuminated stellar nurseries. The brightest regions become intense emission nebulae illuminated by nearby massive stars. Fill the scene with billions of tiny distant stars visible through transparent gas clouds. Use realistic blue, cyan, teal, magenta and golden emission colors based on real astrophotography. Create soft volumetric lighting, multiple layers of gas, subtle light scattering, realistic HDR exposure and immense astronomical scale. The camera performs a smooth cinematic flight through the nebula with gentle inertia and subtle parallax. The environment should resemble real NASA and ESA deep-space images, the Pillars of Creation, the Carina Nebula and the Omega Nebula rather than abstract fractals. Ultra-realistic astrophotography, IMAX documentary, physically based rendering, extremely high detail, temporal consistency, stable geometry, no flickering.
Transform the animation into a breathtaking cinematic journey through the interior of an immense emission nebula. Preserve the original camera path and overall geometry, but reinterpret every fractal surface as colossal volumetric clouds of ionized hydrogen, oxygen and cosmic dust. Replace all solid structures with translucent gaseous formations, turbulent plasma filaments, glowing dust lanes, dense molecular clouds and illuminated stellar nurseries. The brightest regions become intense emission nebulae illuminated by nearby massive stars. Fill the scene with billions of tiny distant stars visible through transparent gas clouds. Use realistic blue, cyan, teal, magenta and golden emission colors based on real astrophotography. Create soft volumetric lighting, multiple layers of gas, subtle light scattering, realistic HDR exposure and immense astronomical scale. The camera performs a smooth cinematic flight through the nebula with gentle inertia and subtle parallax. The environment should resemble real NASA and ESA deep-space images, the Pillars of Creation, the Carina Nebula and the Omega Nebula rather than abstract fractals. Ultra-realistic astrophotography, IMAX documentary, physically based rendering, extremely high detail, temporal consistency, stable geometry, no flickering.
Transform the original object into a colossal futuristic spacecraft while preserving the original camera movement, timing, rotation speed, framing and motion exactly. Reinterpret the entire structure as a gigantic hard-surface starship with layered metallic armor, detailed hull panels, illuminated windows, antennas, sensor arrays, docking bays, engine modules, pipes, vents and complex mechanical greebles. The spacecraft slowly rotates through deep space with majestic scale. Surround it with a realistic starfield, colorful nebulae, faint cosmic dust and subtle lens flares. Illuminate the scene with a distant blue-white star, producing realistic reflections, soft shadows and physically accurate lighting across the hull. Cinematic IMAX science fiction, ultra realistic, photorealistic, high geometric detail, stable temporal consistency, preserve original motion, no deformation, no camera changes, high quality.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (GGUF Q8_0)
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (EVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
241
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.5
Sliding Window
481 klatek
Seed
265489067
Czas generacji
3 min 01 s
Transform the video into a cinematic shot of a massive futuristic spacecraft slowly rotating in deep space. Preserve the original camera movement, timing and motion exactly, but reinterpret the entire structure as an enormous highly detailed starship with metallic hull panels, illuminated windows, antennas, engines, docking bays and intricate mechanical details. Surround it with distant stars, colorful nebulae, faint cosmic dust and subtle lens flares. The spacecraft rotates majestically while drifting through space, illuminated by a distant blue-white star with realistic reflections and shadows. Ultra realistic, hard surface science fiction, photorealistic, physically based rendering, IMAX cinematic, volumetric lighting, extremely detailed, 8K, temporal consistency.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (GGUF Q8_0)
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (EVG)
Rozdzielczość generacji
1280 × 704 px
Liczba klatek
241
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.5
Sliding Window
481 klatek
Seed
105979272
Czas generacji
6 min 08 s
Transform the video into a cinematic shot of a massive futuristic spacecraft slowly rotating in deep space. Preserve the original camera movement, timing and motion exactly, but reinterpret the entire structure as an enormous highly detailed starship with metallic hull panels, illuminated windows, antennas, engines, docking bays and intricate mechanical details. Surround it with distant stars, colorful nebulae, faint cosmic dust and subtle lens flares. The spacecraft rotates majestically while drifting through space, illuminated by a distant blue-white star with realistic reflections and shadows. Ultra realistic, hard surface science fiction, photorealistic, physically based rendering, IMAX cinematic, volumetric lighting, extremely detailed, 8K, temporal consistency.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (GGUF Q8_0)
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
1280 × 704 px
Liczba klatek
241
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
1.0
Sliding Window
481 klatek
Seed
785397225
Czas generacji
8 min 53 s
Transform the video into a cinematic shot of a massive futuristic spacecraft slowly rotating in deep space. Preserve the original camera movement, timing and motion exactly, but reinterpret the entire structure as an enormous highly detailed starship with metallic hull panels, illuminated windows, antennas, engines, docking bays and intricate mechanical details. Surround it with distant stars, colorful nebulae, faint cosmic dust and subtle lens flares. The spacecraft rotates majestically while drifting through space, illuminated by a distant blue-white star with realistic reflections and shadows. Ultra realistic, hard surface science fiction, photorealistic, physically based rendering, IMAX cinematic, volumetric lighting, extremely detailed, 8K, temporal consistency.
Parametr
Wartość
Model AI
VACE 14B
Model bazowy
Wan 2.1 Text-to-Video 14B (INT8)
Typ generacji
Text-to-Video (DSV)
Rozdzielczość generacji
832 × 480 px (480p)
Liczba klatek
81
Liczba kroków (Steps)
20
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
1.0
Sliding Window
81 klatek
Seed
412479861
Czas generacji
25 min 33 s
Transform the video into a cinematic shot of a massive futuristic spacecraft slowly rotating in deep space. Preserve the original camera movement, timing and motion exactly, but reinterpret the entire structure as an enormous highly detailed starship with metallic hull panels, illuminated windows, antennas, engines, docking bays and intricate mechanical details. Surround it with distant stars, colorful nebulae, faint cosmic dust and subtle lens flares. The spacecraft rotates majestically while drifting through space, illuminated by a distant blue-white star with realistic reflections and shadows. Ultra realistic, hard surface science fiction, photorealistic, physically based rendering, IMAX cinematic, volumetric lighting, extremely detailed, 8K, temporal consistency.
Parametr
Wartość
Model AI
VACE 14B
Model bazowy
Wan 2.1 Text-to-Video 14B (INT8)
Typ generacji
Text-to-Video (DSV)
Rozdzielczość generacji
832 × 480 px (480p)
Liczba klatek
81
Liczba kroków (Steps)
20
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
0.6
Sliding Window
81 klatek
Seed
538636037
Czas generacji
25 min 35 s
Transform the video into a cinematic shot of a massive futuristic spacecraft slowly rotating in deep space. Preserve the original camera movement, timing and motion exactly, but reinterpret the entire structure as an enormous highly detailed starship with metallic hull panels, illuminated windows, antennas, engines, docking bays and intricate mechanical details. Surround it with distant stars, colorful nebulae, faint cosmic dust and subtle lens flares. The spacecraft rotates majestically while drifting through space, illuminated by a distant blue-white star with realistic reflections and shadows. Ultra realistic, hard surface science fiction, photorealistic, physically based rendering, IMAX cinematic, volumetric lighting, extremely detailed, 8K, temporal consistency.
Transform the animation into an ultra-realistic microscopic journey inside a gigantic alien cell. Preserve the original camera path and geometry exactly. Every fractal structure becomes intracellular biology: membranes, mitochondria-like energy structures, endoplasmic networks, protein filaments, cytoplasm, giant nuclei, vesicles, molecular machinery, living organelles and flowing biological fluids. The environment is filled with translucent membranes, suspended particles, flowing proteins, bioluminescent enzymes and microscopic fibers. Wet organic materials, realistic subsurface scattering, cinematic macro lens, shallow volumetric fog, extreme biological realism, IMAX nature documentary.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (GGUF Q8_0)
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
1280 × 704 px
Liczba klatek
481
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.9
Sliding Window
481 klatek
Seed
369847473
Czas generacji
13 min 04 s
Transform the animation into an ultra-realistic microscopic journey inside a gigantic alien cell. Preserve the original camera path and geometry exactly. Every fractal structure becomes intracellular biology: membranes, mitochondria-like energy structures, endoplasmic networks, protein filaments, cytoplasm, giant nuclei, vesicles, molecular machinery, living organelles and flowing biological fluids. The environment is filled with translucent membranes, suspended particles, flowing proteins, bioluminescent enzymes and microscopic fibers. Wet organic materials, realistic subsurface scattering, cinematic macro lens, shallow volumetric fog, extreme biological realism, IMAX nature documentary.
rocks, cliffs, crystals, buildings, city, machinery, metal, mechanical parts, concrete, stone, desert, cave, mountain, fantasy landscape, low detail, cartoon
Parametr
Wartość
Model AI
VACE 1.3B
Model bazowy
Wan 2.1 Text-to-Video 1.3B
Typ generacji
Text-to-Video (DV)
Rozdzielczość generacji
832 × 624 px
Liczba klatek
81
Liczba kroków (Steps)
30
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
1.0
Sliding Window
81 klatek
Seed
932963972
Czas generacji
12 min 45 s
Transform the animation into an ultra-realistic microscopic journey inside a gigantic alien cell. Preserve the original camera path and geometry exactly. Every fractal structure becomes intracellular biology: membranes, mitochondria-like energy structures, endoplasmic networks, protein filaments, cytoplasm, giant nuclei, vesicles, molecular machinery, living organelles and flowing biological fluids. The environment is filled with translucent membranes, suspended particles, flowing proteins, bioluminescent enzymes and microscopic fibers. Wet organic materials, realistic subsurface scattering, cinematic macro lens, shallow volumetric fog, extreme biological realism, IMAX nature documentary.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (GGUF Q8_0)
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (EVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
481
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.5
Sliding Window
481 klatek
Seed
283611110
Czas generacji
4 min 42 s
Transform the animation into an ultra-realistic journey through the interior of a gigantic extraterrestrial living superorganism. Preserve the original camera movement, geometry, scale and perspective exactly. Reinterpret every fractal structure as living biological tissue, enormous cellular colonies, bioluminescent membranes, vascular networks, neural fibers, pulsating organs, translucent biofilms, giant spores, microscopic hairs, sensory cilia and complex alien anatomy. The spherical structures become gigantic living cells and biological organs connected by glowing vascular tissue. Fine fractal filaments become nerves, capillaries and microscopic biological structures. Surfaces are soft, moist, semi-translucent and alive with subtle rhythmic motion. Warm green, cyan, turquoise and amber bioluminescence illuminate the organism from within. Dense volumetric atmosphere, floating spores, microscopic particles, realistic subsurface scattering, wet organic materials, physically based rendering, cinematic macro photography, extreme biological detail. Smooth cinematic camera flight, stable motion, realistic inertia, no shaking, no flickering, preserve temporal consistency.
rocks, cliffs, crystals, buildings, city, machinery, metal, mechanical parts, concrete, stone, desert, cave, mountain, fantasy landscape, low detail, cartoon
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (GGUF Q8_0)
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
481
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.9
Sliding Window
481 klatek
Seed
919820487
Czas generacji
7 min 22 s
Transform the animation into an ultra-realistic microscopic journey inside a gigantic alien cell. Preserve the original camera path and geometry exactly. Every fractal structure becomes intracellular biology: membranes, mitochondria-like energy structures, endoplasmic networks, protein filaments, cytoplasm, giant nuclei, vesicles, molecular machinery, living organelles and flowing biological fluids. The environment is filled with translucent membranes, suspended particles, flowing proteins, bioluminescent enzymes and microscopic fibers. Wet organic materials, realistic subsurface scattering, cinematic macro lens, shallow volumetric fog, extreme biological realism, IMAX nature documentary.
rocks, cliffs, crystals, buildings, city, machinery, metal, mechanical parts, concrete, stone, desert, cave, mountain, fantasy landscape, low detail, cartoon
Preserve the original brown dwarf exactly. Only animate the atmosphere while keeping the camera perfectly static. The brown dwarf rotates slowly on its axis. The bright atmospheric bands exhibit subtle differential motion, with hot plasma flowing smoothly around the planet. Convective cloud structures gently evolve, glowing storms slowly circulate, the polar vortex gradually rotates, and faint thermal waves ripple through the upper atmosphere. Subtle volumetric emission from the hottest regions, realistic gas dynamics, physically based lighting, stable planetary geometry, cinematic astrophotography, NASA quality, extremely coherent temporal consistency, no changes to the planet’s overall appearance.
changing planet shape, rocky surface, camera movement, zoom, orbit, morphing, flickering, extra stars, extra planets, deformation, unstable atmosphere
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (FP8)
Model bazowy
LTX-2 19B
Typ generacji
Image-to-Video (Single Image)
Rozdzielczość generacji
1280 × 704 px
Liczba klatek
241
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Input Strength
1.0
Sliding Window
481 klatek
Seed
336583506
Czas generacji
3 min 26 s
Preserve the original brown dwarf exactly. Only animate the atmosphere while keeping the camera perfectly static. The brown dwarf rotates slowly on its axis. The bright atmospheric bands exhibit subtle differential motion, with hot plasma flowing smoothly around the planet. Convective cloud structures gently evolve, glowing storms slowly circulate, the polar vortex gradually rotates, and faint thermal waves ripple through the upper atmosphere. Subtle volumetric emission from the hottest regions, realistic gas dynamics, physically based lighting, stable planetary geometry, cinematic astrophotography, NASA quality, extremely coherent temporal consistency, no changes to the planet’s overall appearance.
changing planet shape, rocky surface, camera movement, zoom, orbit, morphing, flickering, extra stars, extra planets, deformation, unstable atmosphere
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (FP8)
Model bazowy
LTX-2 19B
Typ generacji
Image-to-Video (Single Image)
Rozdzielczość generacji
1280 × 704 px
Liczba klatek
81
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Input Strength
1.0
Sliding Window
481 klatek
Seed
293461708
Czas generacji
2 min 10 s
Preserve the original brown dwarf exactly. Only animate the atmosphere while keeping the camera perfectly static. The brown dwarf rotates slowly on its axis. The bright atmospheric bands exhibit subtle differential motion, with hot plasma flowing smoothly around the planet. Convective cloud structures gently evolve, glowing storms slowly circulate, the polar vortex gradually rotates, and faint thermal waves ripple through the upper atmosphere. Subtle volumetric emission from the hottest regions, realistic gas dynamics, physically based lighting, stable planetary geometry, cinematic astrophotography, NASA quality, extremely coherent temporal consistency, no changes to the planet’s overall appearance.
changing planet shape, rocky surface, camera movement, zoom, orbit, morphing, flickering, extra stars, extra planets, deformation, unstable atmosphere
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled 19B (FP8)
Model bazowy
LTX-2 19B
Typ generacji
Image-to-Video (Single Image)
Rozdzielczość generacji
1280 × 704 px
Liczba klatek
241
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Input Strength
1.0
Sliding Window
481 klatek
Seed
809891378
Czas generacji
3 min 22 s
Preserve the original brown dwarf exactly. Only animate the atmosphere while keeping the camera perfectly static. The brown dwarf rotates slowly on its axis. The bright atmospheric bands exhibit subtle differential motion, with hot plasma flowing smoothly around the planet. Convective cloud structures gently evolve, glowing storms slowly circulate, the polar vortex gradually rotates, and faint thermal waves ripple through the upper atmosphere. Subtle volumetric emission from the hottest regions, realistic gas dynamics, physically based lighting, stable planetary geometry, cinematic astrophotography, NASA quality, extremely coherent temporal consistency, no changes to the planet’s overall appearance.
changing planet shape, rocky surface, camera movement, zoom, orbit, morphing, flickering, extra stars, extra planets, deformation, unstable atmosphere
Parametr
Wartość
Model AI
LTX-2 2.3 Distilled 1.1 22B (INT8)
Model bazowy
LTX-2 22B
Typ generacji
Image-to-Video (Single Image)
Rozdzielczość generacji
1280 × 704 px
Liczba klatek
241
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Input Strength
1.0
Sliding Window
481 klatek
Seed
772378671
Czas generacji
5 min 15 s
Preserve the original brown dwarf exactly. Only animate the atmosphere while keeping the camera perfectly static. The brown dwarf rotates slowly on its axis. The bright atmospheric bands exhibit subtle differential motion, with hot plasma flowing smoothly around the planet. Convective cloud structures gently evolve, glowing storms slowly circulate, the polar vortex gradually rotates, and faint thermal waves ripple through the upper atmosphere. Subtle volumetric emission from the hottest regions, realistic gas dynamics, physically based lighting, stable planetary geometry, cinematic astrophotography, NASA quality, extremely coherent temporal consistency, no changes to the planet’s overall appearance.
changing planet shape, rocky surface, camera movement, zoom, orbit, morphing, flickering, extra stars, extra planets, deformation, unstable atmosphere
Parametr
Wartość
Model AI
Hunyuan Video 1.5 Image-to-Video 8B (480p, INT8)
Model bazowy
Hunyuan Video 1.5 I2V
Typ generacji
Image-to-Video (Single Image)
Rozdzielczość generacji
832 × 624 px
Liczba klatek
129
Liczba kroków (Steps)
30
CFG (Guidance Scale)
6
Flow Shift
5
Sliding Window
129 klatek
Seed
759531441
Czas generacji
28 min 40 s
Preserve the original brown dwarf exactly. A massive brown dwarf slowly rotates in deep interstellar space while maintaining its original appearance and composition. Animate only the atmosphere. Bright orange and crimson cloud bands flow smoothly around the sphere with realistic differential rotation. Hot infrared cloud layers continuously evolve through turbulent convection. A large cyclonic vortex near the north pole rotates slowly while remaining stable. Subtle glowing hot spots pulse beneath semi-transparent clouds, revealing the immense internal heat of the brown dwarf. Soft volumetric thermal emission, realistic atmospheric turbulence, physically accurate gas giant dynamics, cinematic NASA astrophotography, scientific visualization, ultra realistic, highly detailed, stable planetary geometry, temporal consistency, no camera movement, no scene changes.
Preserve the original image exactly. Animate only the atmospheric motion and slow planetary rotation. Keep the overall appearance, cloud pattern, colors and geometry unchanged. A massive brown dwarf floating in deep space, slowly rotating on its axis. Dark reddish-brown atmosphere with glowing orange and crimson cloud bands. Hot infrared cloud layers slowly flow around the planet with realistic differential rotation. A large cyclonic storm near the north pole rotates gradually. Turbulent convection cells continuously evolve, bright hot spots pulse softly beneath semi-transparent clouds. Subtle thermal shimmer, realistic volumetric atmosphere, physically accurate gas giant dynamics, cinematic NASA astrophotography, ultra detailed, stable spherical geometry, smooth continuous motion, deep black space with distant stars, no camera movement, highly coherent animation.
Preserve the original image exactly. Animate only the atmospheric motion and slow planetary rotation. Keep the overall appearance, cloud pattern, colors and geometry unchanged. A massive brown dwarf floating in deep space, slowly rotating on its axis. Dark reddish-brown atmosphere with glowing orange and crimson cloud bands. Hot infrared cloud layers slowly flow around the planet with realistic differential rotation. A large cyclonic storm near the north pole rotates gradually. Turbulent convection cells continuously evolve, bright hot spots pulse softly beneath semi-transparent clouds. Subtle thermal shimmer, realistic volumetric atmosphere, physically accurate gas giant dynamics, cinematic NASA astrophotography, ultra detailed, stable spherical geometry, smooth continuous motion, deep black space with distant stars, no camera movement, highly coherent animation.
Preserve the original brown dwarf exactly. A massive brown dwarf slowly rotates in deep interstellar space while maintaining its original appearance and composition. Animate only the atmosphere. Bright orange and crimson cloud bands flow smoothly around the sphere with realistic differential rotation. Hot infrared cloud layers continuously evolve through turbulent convection. A large cyclonic vortex near the north pole rotates slowly while remaining stable. Subtle glowing hot spots pulse beneath semi-transparent clouds, revealing the immense internal heat of the brown dwarf. Soft volumetric thermal emission, realistic atmospheric turbulence, physically accurate gas giant dynamics, cinematic NASA astrophotography, scientific visualization, ultra realistic, highly detailed, stable planetary geometry, temporal consistency, no camera movement, no scene changes.
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled GGUF Q8_0 19B
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
241
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
1.0
Sliding Window
481 klatek
Seed
423727008
Czas generacji
4 min 46 s
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled GGUF Q8_0 19B
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
1280 × 704 px
Liczba klatek
241
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.3
Sliding Window
481 klatek
Seed
784565988
Czas generacji
6 min 49 s
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled GGUF Q8_0 19B
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
241
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.2
Sliding Window
481 klatek
Seed
466724825
Czas generacji
4 min 24 s
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Parametr
Wartość
Model AI
VACE 1.3B
Model bazowy
Wan 2.1 Text-to-Video 1.3B
Typ generacji
Text-to-Video (DV)
Rozdzielczość generacji
832 × 480 px (480p)
Liczba klatek
161
Liczba kroków (Steps)
30
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
1.0
Sliding Window
81 klatek
Seed
636656649
Czas generacji
20 min 4 s
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Fraktal aexion2, 640×360, 30 fps ↓
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled GGUF Q8_0 19B
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
481
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.3
Sliding Window
481 klatek
Seed
5462476
Czas generacji
7 min 8 s
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled GGUF Q8_0 19B
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
697
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.3
Sliding Window
481 klatek
Seed
807857894
Czas generacji
11 min 21 s
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled GGUF Q8_0 19B
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
737
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.3
Sliding Window
481 klatek
Seed
781339855
Czas generacji
11 min 40 s
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Parametr
Wartość
Model AI
LTX-2 2.0 Distilled GGUF Q8_0 19B
Model bazowy
LTX-2 19B
Typ generacji
Video-to-Video (DVG)
Rozdzielczość generacji
832 × 448 px
Liczba klatek
737
Liczba kroków (Steps)
8
CFG Phase 1
1.0
CFG Phase 2
5.0
Denoising Strength
0.2
Sliding Window
481 klatek
Seed
86877070
Czas generacji
11 min 40 s
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Fraktal aexion2, 832×480, 16 fps, 81 klatek ↓
Parametr
Wartość
Model AI
VACE 1.3B
Model bazowy
Wan 2.1 Text-to-Video 1.3B
Typ generacji
Scene-to-Video (DSV)
Rozdzielczość generacji
832 × 480 px (480p)
Liczba klatek
81
Liczba kroków (Steps)
30
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
1.0
Sliding Window
81 klatek
Seed
106385565
Czas generacji
12 min 18 s
Ultra-smooth cinematic low-altitude flyover above an immense alien megacity carved from surreal fractal architecture. The camera glides only a few meters above the ground, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires rise into the dark sky, with endless districts stretching to the horizon. Soft volumetric fog fills the streets, glowing atmospheric haze surrounds the buildings, mysterious blue and violet energy lights flicker in the distance. The city feels ancient, abandoned, and built by an advanced extraterrestrial civilization. Strong sense of scale, slow cinematic motion, subtle parallax, gentle camera stabilization, realistic inertia, no sudden turns, no zoom, no shaking, preserve the original fractal geometry and textures, highly detailed, atmospheric, science fiction, 8K.
Parametr
Wartość
Model AI
VACE 1.3B
Model bazowy
Wan 2.1 Text-to-Video 1.3B
Typ generacji
Text-to-Video (DV)
Rozdzielczość generacji
832 × 480 px (480p)
Liczba klatek
81
Liczba kroków (Steps)
20
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
1.0
Sliding Window
81 klatek
Seed
355907925
Czas generacji
6 min 39 s
Ultra-smooth cinematic low-altitude flyover above an immense alien megacity carved from surreal fractal architecture. The camera glides only a few meters above the ground, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires rise into the dark sky, with endless districts stretching to the horizon. Soft volumetric fog fills the streets, glowing atmospheric haze surrounds the buildings, mysterious blue and violet energy lights flicker in the distance. The city feels ancient, abandoned, and built by an advanced extraterrestrial civilization. Strong sense of scale, slow cinematic motion, subtle parallax, gentle camera stabilization, realistic inertia, no sudden turns, no zoom, no shaking, preserve the original fractal geometry and textures, highly detailed, atmospheric, science fiction, 8K.
Parametr
Wartość
Model AI
VACE 1.3B
Model bazowy
Wan 2.1 Text-to-Video 1.3B
Typ generacji
Text-to-Video (DV)
Rozdzielczość generacji
832 × 480 px (480p)
Liczba klatek
81
Liczba kroków (Steps)
20
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
1.0
Sliding Window
81 klatek
Seed
813564526
Czas generacji
6 min 40 s
Ultra-smooth cinematic low-altitude flyover above an immense alien megacity carved from surreal fractal architecture. The camera glides only a few meters above the ground, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires rise into the dark sky, with endless districts stretching to the horizon. Soft volumetric fog fills the streets, glowing atmospheric haze surrounds the buildings, mysterious blue and violet energy lights flicker in the distance. The city feels ancient, abandoned, and built by an advanced extraterrestrial civilization. Strong sense of scale, slow cinematic motion, subtle parallax, gentle camera stabilization, realistic inertia, no sudden turns, no zoom, no shaking, preserve the original fractal geometry and textures, highly detailed, atmospheric, science fiction, 8K.
Parametr
Wartość
Model AI
VACE 1.3B
Model bazowy
Wan 2.1 Text-to-Video 1.3B
Typ generacji
Text-to-Video (DLV)
Rozdzielczość generacji
832 × 480 px (480p)
Liczba klatek
81
Liczba kroków (Steps)
30
CFG (Guidance Scale)
5
Flow Shift
5
Sampler
UniPC
ControlNet Weight
1.0
Sliding Window
81 klatek
Seed
123950696
Czas generacji
12 min 21 s
The camera executes a silent, low-altitude glide mere meters above the terrain, moving steadily forward between gigantic crystalline towers and impossible organic structures. Massive cathedral-like spires pierce the dark sky, framing endless districts that stretch to the horizon. Soft volumetric fog blankets the streets, generating a glowing atmospheric haze that envelopes the buildings. Distant blue and violet energy lights flicker rhythmically, casting long shadows across the ancient, abandoned landscape built by an advanced extraterrestrial civilization. Maintain strong sense of scale with slow cinematic motion and subtle parallax shifts. Ensure gentle camera stabilization to preserve realistic inertia. Avoid any sudden turns, zooms, or shaking. Keep the original fractal geometry and textures intact. Render highly detailed, atmospheric science fiction visuals in crisp 8K resolution.
Podsumowanie testów modeli AI w WanGP
Po wykonaniu kilkudziesięciu testów na stacji roboczej HP Z440 z RTX 5060 16 GB można wyciągnąć następujące wnioski.
LTX-2 (19B / 22B)
To obecnie najlepszy model do generowania wideo.
Zalety:
zdecydowanie najszybszy spośród testowanych modeli,
najwyższa jakość obrazu,
bardzo dobra spójność między klatkami,
obsługa bardzo długich sekwencji (nawet 737 klatek),
duże Sliding Window – 481 klatek,
obsługuje zarówno Image-to-Video, jak i Video-to-Video,
możliwość generowania filmu wraz z dźwiękiem.
Profile pamięci:
Image-to-Video 720p – najlepiej używać Profilu 3, zużycie około 13–15 GB VRAM.
Video-to-Video – bez problemu działa na Profilu 2.
Wan VACE
Model bardzo dobrze sprawdza się jako narzędzie Video-to-Video z zachowaniem ruchu oraz geometrii oryginalnego filmu.
Zalety:
bardzo dobre efekty stylizacji,
świetne zachowanie ruchu kamery,
dobre wykorzystanie ControlNet.
Wady:
wyraźnie wolniejszy od LTX-2,
maksymalnie około 81–121 klatek,
znacznie mniejsze Sliding Window (81–121 klatek).
Najlepiej sprawdza się przy krótkich transformacjach materiałów źródłowych.
Hunyuan Video 1.5
Model oferuje dobrą jakość Image-to-Video, jednak wymaga odpowiedniego doboru profilu.
Najlepsza konfiguracja:
Image-to-Video 480p
Profil 3
około 13 GB VRAM
Przy użyciu Profilu 2 model działał mniej optymalnie.
Profile pamięci WanGP
Profil 2
preferowany dla modeli z dużym offloadem do RAM,
bardzo niskie zużycie VRAM (np. Wan Image-to-Video około 6 GB VRAM),
model pozostaje w pamięci RAM i do VRAM trafiają tylko potrzebne fragmenty.
Profil 3
większe wykorzystanie VRAM (około 13–16 GB w zależności od modelu),
zalecany dla bardziej wymagających modeli, takich jak:
LTX-2 Image-to-Video (720p+),
Hunyuan Image-to-Video,
warto przełączyć się na Profil 3, gdy Profil 2 powoduje błędy(OOM RAM).
Jeżeli pojawi się błąd Out of Memory (OOM), warto poeksperymentować z przełączeniem między Profilem 2 i 3, ponieważ skuteczność zależy od konkretnego modelu i sposobu jego offloadu.
Najważniejszy wniosek: obecnie LTX-2 oferuje najlepszy kompromis między szybkością, jakością obrazu i długością generowanych sekwencji. Wan VACE pozostaje bardzo dobrym wyborem do stylizacji istniejących filmów (Video-to-Video), natomiast Hunyuan 1.5 sprawdza się głównie w Image-to-Video przy odpowiednio dobranym profilu pamięci.
Ta strona korzysta z ciasteczek, aby zapewnić Ci najlepszą możliwą obsługę. Informacje o ciasteczkach są przechowywane w przeglądarce i wykonują funkcje takie jak rozpoznawanie Cię po powrocie na naszą stronę internetową i pomaganie naszemu zespołowi w zrozumieniu, które sekcje witryny są dla Ciebie najbardziej interesujące i przydatne.
Ściśle niezbędne ciasteczka
Niezbędne ciasteczka powinny być zawsze włączone, abyśmy mogli zapisać twoje preferencje dotyczące ustawień ciasteczek.
Jeśli wyłączysz to ciasteczko, nie będziemy mogli zapisać twoich preferencji. Oznacza to, że za każdym razem, gdy odwiedzasz tę stronę, musisz ponownie włączyć lub wyłączyć ciasteczka.
Dodaj komentarz