Turn a short creative brief into a video and guide it with visual, motion, or audio references. H3 Max is CawCut’s fast-generation option in the H3 family, with text, image, frame, and omni-reference modes in the [Video Generation] node.
1. H3 Max Strengths
| Strength | Application |
|---|---|
| Fast generation with strong quality | Explore short video ideas and iterate without giving up the model’s quality-focused profile. |
| Prompt understanding and aesthetics | Use it for scenes where the visual style and faithful reading of the brief both matter. |
| Multimodal reference workflows | Combine text with supported image, video, or audio references to direct the result. |
| Flexible scene creation | Use text-only concepts, image animation, first/last frames, or omni-reference setups. |
2. Modes and Inputs
H3 Max supports Text, Image References, Frames, and Omni References in CawCut.
| Mode | Use it for | Typical input |
|---|---|---|
| [Text] | Creating a scene from a written brief. | Prompt describing subject, action, setting, and camera. |
| [Image References] | Animating or guiding an existing visual. | Image or character reference with a motion direction. |
| [Frames] | Controlling the opening and optional ending composition. | First frame, optional last frame, and transition direction. |
| [Omni References] | Coordinating several media types in one generation. | Supported images, character, video, or audio references plus a prompt. |
3. Prepare Prompts and References
Explain what the viewer should see first, what changes, and how the shot should finish. Assign each reference one role—subject appearance, motion, camera, sound, or visual style—and state the action in clear chronological order.
What to avoid: Avoid contradictory reference material, vague movement verbs, and a long list of unrelated visual details. If a specific action matters, describe its start, movement, and end state.
For a clear prompt:
- Identify the main subject and the key action.
- Put events in the order they should happen.
- Give the camera one coherent direction.
- State which subject or scene details should remain consistent.
- Add style and audio direction only when they support the intended result.
4. CawCut Workflow and Settings
- Add a Video Generation node and select Text, Image References, Frames, or Omni References.
- Choose H3 Max and provide the input references supported by the mode.
- Write the core action first, then specify camera movement, visual treatment, and audio direction as needed.
- Set the available duration, aspect ratio, resolution, and audio options.
- Review the credit estimate and generate. Watch the entire clip for motion, sound, and consistency.
- Refine the prompt or references based on the most important mismatch.
For full mode and setting details, see Video Generation. For connecting this node with other steps, see Understanding Nodes.
5. Limitations and Refinement
A strong prompt and references improve direction but do not guarantee exact motion or timing. Multi-subject interactions and dense sequences can still vary; simplify the action or split a complex idea into separate clips when needed.
| If you notice | Try this |
|---|---|
| The action is hard to follow | Break it into ordered beats and focus on one key movement. |
| The subject or visual style drifts | Use a clearer reference and list the details to preserve. |
| Camera movement is confusing | Simplify it to one camera direction and one main action. |
| A detail is almost right | Keep the successful inputs and refine only that element. |
Review the full clip before using it as final footage. Check Credits before running many variations.