CivArchive
    Story Creator 12GB — script to a full story video: Qwen Image 2.1 stills + MiniMax H3 scenes with voices + YuE2 music - v1.4.0
    NSFW
    Preview 148904502
    Preview 148904498
    Preview 148904511
    Preview 148904911
    Preview 148904521
    Preview 148905858

    ☕ One coffee = one happy creator: https://buymeacoffee.com/the_frizzy1


    A script goes in, a whole story video comes out: the same character in every scene, every scene moving, characters talking on screen, a narrator over the top, music. The long AI story video trend, run on your own PC. Qwen Image 2.1 Turbo makes the stills, MiniMax H3 turns each one into a moving scene with sound, YuE2 makes the song.

    What you need

    ✔ Custom nodes — None. Core ComfyUI only.
    ✔ ComfyUI version — 0.39 or newer (Start Loop / End Loop, MiniMax H3, YuE2). Older versions in source/.
    ✔ VRAM — 12 GB — tested, not estimated
    ✔ System RAM — 32 GB. ComfyUI's memory grows about 0.6 GB per scene.
    ✔ Tested on — RTX 3060 12 GB · 32 GB DDR4 · ComfyUI 0.39.2
    ✔ Licence — Workflow MIT · Qwen Image 2.1: Qwen Research Licence (non-commercial) · MiniMax H3: MiniMax H3 Community Licence · YuE2: CC-BY-NC-4.0 (non-commercial)

    Models to download

    Click a file name to download it. Put it in the folder shown, inside ComfyUI/models/. Nine files in all: 3 for the stills, 5 for the video, 1 for the music.

    Stills — Comfy-Org/Qwen-Image-2.1

    diffusion_models/

    File: qwen_image_2.1_turbo_int8_convrot.safetensors — 7.3 GB (stills)
    Download:
    https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_turbo_int8_convrot.safetensors

    Put it here:
    ComfyUI/models/diffusion_models/

    text_encoders/

    File: qwen3vl_8b_int8_convrot.safetensors — 9.4 GB (stills)
    Download:
    https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_int8_convrot.safetensors

    Put it here:
    ComfyUI/models/text_encoders/

    vae/

    File: qwen_image_2.1_vae_bf16.safetensors — 0.7 GB (stills)
    Download:
    https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/vae/qwen_image_2.1_vae_bf16.safetensors

    Put it here:
    ComfyUI/models/vae/

    Video — Comfy-Org/MiniMax-H3

    diffusion_models/

    File: minimax_h3_fl2va_pruned_int8_convrot.safetensors — 21.0 GB (video)
    Download:
    https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors

    Put it here:
    ComfyUI/models/diffusion_models/

    loras/

    File: minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors — 2.0 GB (video)
    Download:
    https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/loras/minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors

    Put it here:
    ComfyUI/models/loras/

    text_encoders/

    File: qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors — 15.7 GB (video)
    Download:
    https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors

    Put it here:
    ComfyUI/models/text_encoders/

    vae/

    File: minimax_h3_video_vae_int8_convrot.safetensors — 2.8 GB (video)
    Download:
    https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_video_vae_int8_convrot.safetensors

    Put it here:
    ComfyUI/models/vae/

    File: minimax_h3_audio_vae_fp32.safetensors — 0.6 GB (video)
    Download:
    https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_audio_vae_fp32.safetensors

    Put it here:
    ComfyUI/models/vae/

    Music — Comfy-Org/YuE2

    checkpoints/

    File: yue2_3b_int8_convrot.safetensors — 4.0 GB (music)
    Download:
    https://huggingface.co/Comfy-Org/YuE2/resolve/main/checkpoints/yue2_3b_int8_convrot.safetensors

    Put it here:
    ComfyUI/models/checkpoints/

    The script format

    WORLD: A snowy northern land in winter, soft blue and pink light, cozy and magical
    [wide, slow push-in] Inside a cozy den by the fire: Pip, a small orange fox in a knitted blue scarf, sits next to Grandpa, an old grey fox with a walking stick || In a den under a snowy hill lived a little fox named Pip.
    [medium shot] Inside the cozy den: Pip, the small orange fox in the blue scarf, jumps up and points at a painting of green lights on the wall || Pip: Grandpa, are the northern lights real?
    [continue] [wide shot] Morning outside the round wooden door of the den: Pip waves goodbye, holding a small wooden compass || So the next morning, Pip set off with the compass.

    WORLD: sets the look of the whole story · [camera] is the shot and the camera move · then what we see: place, every character with their full look, objects, the action · || text the narrator says off screen · || Name: text that character says on screen · [continue] ends the clip on the next scene's picture instead of cutting.

    Scene length is automatic, 3-9 s, from how much is said. The chatbot prompt and the rules that make the pictures match the words are in HOW_TO_WRITE_A_SCRIPT.md. A 58-line example, about 5 minutes: samples/example_script_5min.txt.

    How it works

    Music first, if it is on: YuE2 makes a song and every scene later gets its own slice. Then Qwen Image 2.1 Turbo makes one character picture and every scene's still as an edit of it, which is why the hero stays the same. Then a Start Loop / End Loop walks the stills: MiniMax H3 with the 8-step turbo LoRA turns each one into a clip with sound, as long as its line needs. Narrator lines get their own small H3 pass and only its sound is used, so one voice carries the whole story. Each scene is saved as it finishes, then they are joined.

    Times on my RTX 3060 (0.4 MP)

    • 1 min (v1.0, turbo LoRA, 5 s scenes) — 12 scenes, 43 min 38 s time

    • 3 min (v1.0, turbo LoRA, 5 s scenes) — 36 scenes, 2 h 05 min time

    • 5 min (v1.2, FastH3, no upscale) — 58 scenes, 3 h 35 min time

    Most of that is H3. A 5 minute story is an overnight job, not a coffee break.

    What did not work

    • The built-in writer is off for a reason. "Write the scenes for me", where the 8B Qwen text encoder writes the scene list from an idea, is still in the workflow but switched off. Its lines came out short and vague and it dropped how side characters look — a white rabbit became a second fox. A detailed hand-written script got 10 of 12 scene stills right first try. Write it with a big chatbot instead.

    • Only the main character stays the same, because there is one character picture. Side characters live in your words only, so describe them fully in every line they are in.

    • Over-the-shoulder shots and characters cut off at the frame edge go wrong: the hero came out twice, or two characters melted together.

    • The loudness differs per scene, because H3 makes each scene's sound separately. Level it in your editor.

    • Over about 30 scenes, run the script in two halves and join them in your editor. ComfyUI's memory grows about 0.6 GB per scene on a 32 GB PC.

    Get the workflow

    More videos on this stuff: https://www.youtube.com/playlist?list=UUGKDBMhTGqvlBZE9cF5d50Q


    Description

    v1.4.0. The README has no changelog section, so I cannot say what changed. The timing rows on this page are from v1.0 and v1.2 runs, not from v1.4.

    Workflows
    Other

    Details

    Downloads
    14
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/20/2026
    Updated
    11/20/2026
    Deleted
    -

    Files

    The_frizzy1_story-creator-12gb_v1.4.0.json