Kai opens a washing machine. Mochi gets a face full of foam. Niko cannot stop laughing. Here’s how I build this ten-second anime scene from three character sheets, one location and a shot-by-shot prompt.
Start with the references if you want to animate right away.
The image stage uses my custom Krea 2 Wulver fine-tune. Model weights are not included. The approved PNGs let you move directly to the video stage.
FastH3 model files and settings
Diffusion model: fasth3_omniref_8step_pruned_int8_convrot.safetensors Text encoder: qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors Video VAE: minimax_h3_video_vae_int8_convrot.safetensors Audio VAE: minimax_h3_audio_vae_fp32.safetensors
Use the supplied graph’s model download metadata where available. Resolution Selector #115 is 1:1, 1 MP, multiple 32. Duration #132 is 10 seconds; the 17k + 5 expression requests 243 frames at 24 fps, approximately 10.125 seconds. The graph keeps a fixed seed, comfy kitchen attention, and the original VSA block-sparse settings.
Load the JSON in ComfyUI. Use Manager to install missing nodes, then restart and refresh.
Select your installed compatible models. Use a current ComfyUI build with the FastH3-required H3, ModelAttentionBackend and BlockSparseAttention nodes. The replacement video graph uses core nodes and has no latent upscale stage.
Upload the four PNGs and select them in the Load Image nodes. Keep Kai, Niko, Mochi and the laundromat in that order.
Check duration and resolution before running. The adapted JSON sets the duration control to 10 and replaces prompt #138 with the exact prompt below.
This laundry adaptation is based on FastH3_Omni_8Step_Ref2Video_16GB(1).json. It replaces the car prompt, sets ten seconds and 1:1 output, and connects four references in order: Kai, Niko, Mochi, location. It preserves Euler, simple scheduling, 8 steps, denoise 1, video/audio sigma shift 12/3, and the supplied attention settings. Local input files must be relinked. This adaptation has been structurally checked; no new GPU inference was performed. The file’s 16GB label is not a new hardware benchmark.
02 / THE FIELD GUIDE
SAME CAST. NEW ORDINARY MOMENT.
Generate each character separately, then approve the sheet before moving into animation.
KAIPortrait · front · back
Kai · image promptRead + copy ↗
Character sheet of a cheerful 30 year old young Asian man, wearing thick black framed glasses, slim build, messy short black hair, large expressive dark eyes, rosy cheeks. Pale blue T-shirt, black pants, blue Converse sneakers with white toe caps and laces. No apron or overalls. Three views arranged horizontally: LEFT three-quarter close-up portrait, MIDDLE full-body front, RIGHT full-body back. Consistent proportions and outfit, feet fully visible, plain cream background. Hand-drawn 2D anime, chunky organic outlines, expressive cartoon face, painterly cel shading, textured handmade brushwork, vibrant teal-turquoise-orange accents, animated-film concept art. No text, photorealism, or 3D.
NIKOPortrait · front · back
Niko · image promptRead + copy ↗
Character sheet of a cheerful young 28 year old Asian woman, long messy dark-brown hair with bangs, large expressive dark eyes, rosy cheeks. Red T-shirt, teal blue-green denim overalls with rolled cuffs, faded red Converse sneakers with white toe caps and laces. No apron. Three views arranged horizontally: LEFT three-quarter close-up portrait, MIDDLE full-body front, RIGHT full-body back. Consistent proportions and outfit, feet fully visible, plain cream background. Hand-drawn 2D anime, chunky organic outlines, expressive cartoon face, painterly cel shading, textured handmade brushwork, vibrant teal-turquoise-orange accents, animated-film concept art. No text, photorealism, or 3D.
MOCHIPortrait · front · back
Mochi · image promptRead + copy ↗
Wide character sheet of a cute chubby orange tabby cat: round head, plump compact feline body, very short legs, small black oval eyes, pink nose, cream muzzle and chest, darker orange stripes, thick curved tail, teal collar with a small teal locket. Three clearly separated views with generous space: LEFT three-quarter head-and-shoulders close-up; MIDDLE full-body front, sitting naturally with front paws together; RIGHT full-body back in the same seated pose, tail curled beside the body. Small ears, rounded cheeks, consistent proportions. Plain cream background. Hand-drawn 2D anime, chunky organic outlines, painterly cel shading, textured handmade brushwork, warm orange and teal palette. No human anatomy, upright human stance, long limbs, overlapping figures, clothing, text, or 3D.
Give every view the same identity
Kai’s glasses and blue shirt, Niko’s teal overalls, and Mochi’s stripes and collar should agree across views. These are alternate views of one individual. They are not extra people or cats to include in the scene.
Use the Wulver image workflow
Load the custom model, Krea 2 text encoder and VAE. Enter the positive prompt, choose latent dimensions, sample, decode and save. The starter uses 8 steps, CFG 1.0, Euler and simple. Leave its negative prompt empty at this CFG. Approved Kai and Niko sheets are 1280 × 1280; Mochi is 1920 × 1080.
03 / THE FIELD GUIDE
BUILD THE LAUNDROMAT. LEAVE ROOM FOR THE JOKE.
The location establishes the machines, lighting and palette before the cast arrives.
The circular washer doors give Kai a clear action. The tiled floor gives Mochi somewhere to stand and the water somewhere to spread. The top of the adjacent machine gives Niko a distinct seat. Use the same image workflow to make this environment; the approved location reference is 1920 × 1080.
04 / THE FIELD GUIDE
ONE SMALL ACCIDENT. FIVE READABLE SHOTS.
Each shot has a separate job: action, impact, reaction, laughter and payoff.
01 / 0–2s · INTENDED BEAT
OPEN THE DOOR.
Kai opens the washer. Keep Mochi on the floor outside its circular door.
02 / 2–4s · INTENDED BEAT
HERE COMES THE FOAM.
Water and suds travel from the open washer onto Mochi. The cat never emerges from the drum.
03 / 4–6s · INTENDED BEAT
HOLD THE GRUMPY LOOK.
Wet fur, narrowed eyes, flattened ears. A head shake scatters bubbles before the stare lands.
04 / 6–8s · INTENDED BEAT
LET NIKO REACT.
A separate closeup gives her laughter room to read. One hand covers her mouth.
05 / 8–10s · INTENDED BEAT
BRING EVERYONE TOGETHER.
Embarrassed Kai, annoyed Mochi, laughing Niko. A popping bubble and tiny meow finish the joke.
These stills are from the finished clip. The schedule describes intended direction; generated cut timing can vary. This is a multi-shot scene with hard cuts and static cameras within each shot.
05 / THE FIELD GUIDE
REFERENCES IN. A PERFORMANCE OUT.
Watch the mockup, then run the downloadable ComfyUI graph.
The animated mockup illustrates reference ordering and prompting. The downloadable FastH3 graph is the implementation to follow.
Four Load Image nodes Kai → Niko → Mochi → location
→
Reference prompt + MiniMax H3 Identity, action, staging and cuts
→
Sample → decode → video Review the performance
Keep each reference’s role explicit in the prompt. Connect the four images to the graph’s reference inputs and paste the director’s prompt into Input Text (Prompt) #138. Check the duration control at node #132: the laundry scene targets ten seconds. The animation above is a visual explanation of the process; it does not perform inference.
06 / THE FIELD GUIDE
WRITE THE CAUSE. THEN THE REACTION.
The most important correction is spatial: Mochi starts outside the machine and stays there.
Establish positions before action
Kai crouches at the door. Mochi stands on the floor in front of it. Niko sits on the neighboring washer. This staging makes the splash readable and prevents the model from treating the cat as something being unloaded from the drum.
Describe the direction of the water
The door opens; foam bursts outward; the splash lands on Mochi. Repeat the outside-the-washer constraint in the setup, splash shot and final animation directions.
Separate action from expression
The splash gets its own closeup. Mochi’s grumpy stare gets another. Niko’s laughter follows. Asking for all three reactions at once can make the edit feel rushed.
Keep the visual cadence distinct from export fps
The prompt’s 15 fps describes the stepped anime look. The supplied finished clip is encoded at 24 fps. Changing export fps alone changes playback timing; it does not create the intended drawing cadence.
The exact MiniMax H3 laundry promptRead + copy ↗
Create a 10-second static medium closeup shot in a 1990s hand-drawn Japanese anime cel style at 15 fps, with limited frame-by-frame animation and slightly stepped motion.
References:
- <Picture 1>: Kai, messy black hair, round glasses, light blue shirt and black pants.
- <Picture 2>: Niko, long dark hair, red shirt and teal overalls.
- <Picture 3>: Mochi, orange tabby cat with teal collar.
- <Picture 4>: Japanese laundromat with turquoise washing machines, fluorescent lighting and rainy windows.
Scene: A funny, playful moment inside a Japanese laundromat at night. Kai crouches in front of a washing machine filled with excessive soap foam. Mochi stands on the tiled floor directly in front of the closed washer, looking curiously at its circular door. Niko sits on top of a neighboring washing machine, watching everything unfold.
Shot 1 (0-2s) - Medium closeup: Kai crouches beside Mochi in front of the turquoise washer. Mochi stands on the floor, looking curiously at the circular door. Kai reaches forward and opens the washer. A large amount of white soapy water and foam immediately bursts out toward Mochi.
Shot 2 (2-4s) - Closeup: A huge wave of fluffy white soap suds and bubbly water spills from the open washing machine, splashing directly onto Mochi. Mochi stays on the floor in front of the machine as the foam completely soaks his fur. His face becomes covered in bubbles, and he looks startled and comically displeased.
Shot 3 (4-6s) - Closeup: Mochi sits in the puddle of foam with an adorably annoyed expression, narrowed eyes and flattened ears. His wet orange fur is covered in fluffy white bubbles. He shakes his head, scattering bubbles everywhere, then gives Kai an unimpressed, grumpy stare.
Shot 4 (6-8s) - Medium closeup: Niko sits on top of the neighboring washing machine, her legs dangling. She bursts into laughter at the sight of Mochi, covering her mouth with one hand.
Shot 5 (8-10s) - Medium three-character shot: Kai crouches beside the open washer, looking embarrassed. Mochi sits in front of him, soaked and covered in foam, looking playfully annoyed. Niko laughs from her seat on the adjacent washer. A bubble on Mochi's head pops, and he gives a tiny irritated meow.
Animation: Traditional 1990s hand-drawn Japanese anime, 2D cel shading, painted backgrounds, subtle film grain, expressive faces, limited 15 fps animation. Static cameras, hard cuts between shots. Keep characters large in frame. Preserve character designs and spatial continuity. Mochi must remain outside the washing machine throughout the entire scene. The foam and water splash onto him when Kai opens the door. No 3D, CGI, morphing, camera movement or distant wide shots.
Audio: Washing machine hum, door click, sudden rushing soapy water, splashing foam, popping bubbles, Niko's laughter and Mochi's annoyed meow. Soft rain outside. No dialogue or music.
Use Run, watch the whole clip, and change one problem at a time.
01
Check inputs
Confirm all four references, model loaders, ten-second duration and the square output setting. Keep the original sampler path intact while establishing a working baseline.
02
Watch positions and cuts
Mochi must remain outside the machine. Check the splash direction, Kai’s glasses, Niko’s outfit and the order of reactions. The supplied clip is 736 × 736 and approximately 10.13 seconds; target duration and encoded duration can differ.
03
Refine one detail
Record your seed and save the graph before revising. Simplify unclear action before adding detail. Upscaling can improve appearance but will not reliably repair staging or continuity.
08 / THE FIELD GUIDE
KEEP THE CHARM. FIX THE CONFUSION.
Review the result against the intended actions, rather than the prompt alone.
Mochi comes out of the washer.
Replace the entire older laundry prompt. The revised setup places him on the floor before the door opens and explicitly keeps him outside throughout. Check that no earlier “discovers the cat inside” instruction remains.
The model renders multiple cats or people.
Keep one approved reference per character. Explain that the sheet panels describe views of the same individual. Ask for exactly one Kai, one Niko and one Mochi if duplication persists.
The scene becomes one moving camera shot.
Keep the five shot headings, static-camera instruction and hard-cut direction. Remove conflicting continuous-shot or camera-movement directions from earlier prompts.
The reaction disappears in the splash.
Reduce secondary movement. Give the annoyed expression a settled hold after the head shake, then cut to Niko.
The workflow cannot find a model.
Relink installed files in each loader. The downloads include workflow recipes and references, not model weights. The custom Wulver checkpoint is not bundled.
The render runs out of memory.
Try a shorter test with a matching shorter action prompt and a supported smaller canvas. Review the requirements of the exact weights and quantization you installed. No new hardware benchmark was performed for this guide.
EXTRA:ORDINARY LIFE
KEEP THE CAST. FIND THE NEXT MOMENT.
A familiar world, one simple action, and a reaction worth watching.
