About this version

chan log: 1.0a

didnt expect anyone to DL so made more detailed notes

+ more detailed notes in Wf

+fixed on wire from RefMOD to Continuum Sampler that might have been wrong

+added instructions in post

DO THE THING

1. Load the video model

The model-loading boxes supply MiniMax H3 and the supporting parts needed to generate video. Leave your working model selections alone unless you’re deliberately changing models.

2. Load your character RefMod

In box 10b, Load H3 RefMods, select the character file you baked. Your saved annotated copy had “fire girl” selected.

The RefMod helps preserve the character’s appearance. It isn’t a picture reference or a LoRA.

Box 10a, RefMod Continuum Bridge, applies that RefMod to the model. Its output must reach the model input on box 21, the big Continuum Sampler. That’s the connection I fixed in the saved annotated copy.

3. Load your reference pictures

The boxes labeled Reference Image 1 and Reference Image 2 supply pictures you can mention in the prompt:

• Reference Image 1 becomes @R1.

• Reference Image 2 becomes @R2.

• ect ect max 9 ref images (can also ref video, sound, others)

In your saved example, those were the woman and the dragon. These are reference pictures—not the first frame of the movie.

4. Choose when the pictures are used

In box 18, Reference Images, choose whether each picture applies to all chunks or only selected chunks.

For example, if the dragon should guide chunks 1 and 2, enable its image for those chunks. Writing @R2 in the prompt doesn’t turn its checkbox on.

5. Optional first and last frames

The separate FF. FIRST FRAME and LF. LAST FRAME boxes are deliberately disconnected.

Use these when you want to supply an actual starting or ending image, rather than just a visual reference. Their image outputs connect to the corresponding first_frame and last_frame inputs on box 21. Leave them detached when you don’t need them; the pictures currently in them are placeholders.

6. Write your scene

In the prompt box, describe your characters, setting, actions, camera movement, and sound. Use @R1 and @R2 to identify the pictures.

For a chunk-based script, use Timeline format and separate sections such as [Chunk 1], [Chunk 2], and [Chunk 3]. READ PROMPT GUIDE BY OG

7. Set the video in box 21

Box 21 is the actual Continuum Sampler. Box 12 is only the sampling-method selector.

The main settings you’ll use are:

• Chunks and Seconds per Chunk: determine duration. Three chunks at seven seconds gives 21 seconds.

• Prompt Format: use Timeline for your numbered chunk script.

• Generation Mode: keep Full Run / Generate Full Video to generate everything automatically.

• Seed: fixed keeps the same starting seed; changing it gives a different variation.

• Audio Continuity: carries sound between chunks. Describe the sound you want in the prompt too.

The other advanced controls can stay at your working settings while you learn these.

8. Run and save

Once the model, RefMod, pictures, prompt, and duration are set, press Run. The output section saves the generated video.

For everyday use, you mainly change the character RefMod, reference pictures, prompt, and duration. The long model-and-connection chain usually stays as it is.

prompt guide https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum#quick-start--a-different-prompt-for-each-chunk