Character and Scene Assets: Making the Character Look Like the Same Person in Every Shot
💡 Goal: Deliver a complete character and scene asset package within 45 minutes—a setting card plus a set of multi-angle reference images for each character, one reference image for each scene, and along the way finish the style consistency check and asset naming and archiving. Do this AI character and scene step solidly, and the next stop, shot generation, will have images to rely on, so the character looks like the same person in every shot.
🎯 Stage Goals
- Deliver Asset One: Character setting card—fix each character's appearance, clothing, and personality into one fixed description that all subsequent prompts reference
- DeliverAsset Two: Character reference images—a set of multi-angle reference images for each character, with even lighting and a neutral expression, used to identify the character during shot generation
- DeliverAsset Three: Scene reference images—one finalized reference image per scene, used directly as the first frame of image-to-video; the whole asset package passes the style consistency check and is archived by shot number from the storyboard
⏳ Estimated Time
- The five lessons in this stage total about 32 minutes (6 + 8 + 6 + 7 + 5), or about 22–42 minutes allowing ±30% variance
- Short-on-time route: 3.1 Character setting card, 3.2 Character reference images, and 3.3 Scene reference images are must-dos (about 20 minutes); save 3.4 Style consistency for a comparison check after the first shot is generated, and add 3.5 Naming management before batch retries
- Starting from this stop you will generate images: the local route needs a machine with ComfyUI installed, and the cloud route needs a platform account (for how to choose a route, see 1.3 Two Production Routes)
🚀 Before You Start
- Completed Story and Storyboard: especially 2.4 the Storyboard—only with a finalized shot list do you know how many characters and scenes to prepare
- Chosen a production route: 1.3 Cloud tools or local deployment
- Local route: ComfyUI installed and models in place—start from ComfyUI Quick Start; for where to put model files, see Models Folder Guide
- Cloud route: the account for the platform you use has been activated
- All characters generated with AI: this course does not use real people's portraits as characters—using a real person's likeness requires their authorization, so the practice film simply avoids this pitfall
- A cup of water ☕—from this stop on you'll be waiting for images to render, a perfect time to drink water
📚 Lessons in This Stage
The AI character and scene coursework answers five questions in order: What should the character description say? How do you generate reference images? How do you finalize scene images? Is the style consistent across the whole film? How should assets be stored so nothing gets lost?
| Lesson | Content | Duration |
|---|---|---|
| 3.1 Character setting card | Lock down appearance, clothing, and personality: produceAsset One, with the whole film referencing the same description | 6 minutes |
| 3.2 Character reference images | Multi-angle generation method: produceAsset Two, with even lighting and a neutral expression | 8 minutes |
| 3.3 Scene reference images | Set the location before rolling: produceAsset Three, used directly as the first frame of image-to-video | 6 minutes |
| 3.4 Style consistency | Make the whole film feel like one world: style checklist + unified style prompt template | 7 minutes |
| 3.5 Asset naming and version management | Never lose a shot again: archive assets by shot number from the storyboard | 5 minutes |
How the asset package maps to the five lessons (this course's assignment design):
| Delivered Asset | Produced in Which Lesson | Refined in Which Lesson |
|---|---|---|
| Asset One: Character setting card | 3.1 | 3.2 Character reference images (verifying during generation that the description is sufficient) |
| Asset Two: Character reference images | 3.2 | 3.4 Style consistency |
| Asset Three: Scene reference images | 3.3 | 3.4 Style consistency |
| Asset archiving conventions | 3.5 | — |
The entire course is built around the same 30–60-second original practice short film (this is the course design of this site). The previous stop, Story and Storyboards, delivered three paper documents: the synopsis, the script, and the storyboard table; this stop turns the paper into images; the next stop, Shot Generation, uses the reference images to produce the film shot by shot.
This section only covers "what to prepare for a short drama's characters and scenes, and how to check them"; the actual image-generation operations (ComfyUI installation, model placement, text-to-image workflows) reuse the already-published tutorials in the comfyui section, with the corresponding links provided in each lesson where needed.
💡 Why prepare reference images first, then generate shots?
The official workflow itself is "generate multiple short segments, then edit and combine them" (see 1.1 Full Workflow Overview) — each shot is generated separately, and even the same character description can produce two different faces.
The Runway Gen-4 official help documentation recommends: use reference images to guide the characters, objects, or style in the frame, choose input images with even lighting and neutral expressions, then adjust iteratively step by step. This section breaks that idea into three steps: "setting card → reference images → style check" (the teaching design of this site).
The boundary should also be made clear: reference images improve consistency, but they are not a guarantee of "the face never changing across shots" — residual drift in face shape and clothing is caught as a fallback by 2.5 Shot continuity checking and 4.3 First-and-last-frame control.
🎒 What you will gain
- A reusable set of character assets: setting card + multi-angle reference images, so every shot generation cites the same set of references
- A scene reference image that can be used directly as the first frame for image-to-video: when you get to Shot Generation, you won't have to invent the frame on the spot
- A consistency habit of "feeding a reference image + the same description for every shot": don't count on the model remembering the previous shot
- An asset library archived by storyboard shot numbers: no matter how many retries, you can always find the latest version