To make a brand mascot with AI that looks the same everywhere, don't start with a video prompt. Design the character as a still reference sheet first (front, side, back and a few expressions), approve it like a logo, and then make every later image and clip from that sheet as a reference, never from a text description alone. On 8frame the whole process sits on one board: image models that accept 10 to 14 reference images (from 6 credits an image), ShotGrid for a nine-frame sheet in one image (42 credits), and video models that take the sheet as input, such as Happy Horse with 1 to 9 character references (95 credits for 5 seconds at 720p). The part most guides skip: under current US Copyright Office guidance, a mascot made from prompts alone is not protected by copyright, so plan your own edits to it and lean on trademark use.
TL;DR
- A text description is not a character. Two prompts with the same words give two different mascots. The thing that keeps a mascot stable is an approved image you feed back in every time.
- Order of work: brief, 20 to 40 candidate sketches, one chosen design, a reference sheet, a test set of poses, then video. Each step reuses the previous step's output as a reference.
- Reference limits on 8frame image models: Seedream 5 Lite up to 14 (6 credits an image), Nano Banana 2 up to 10 (11 credits at 1K), GPT Image 2 up to 10 (20 to 29 credits at its default high quality).
- For clips: Happy Horse reference-to-video takes 1 to 9 character references (95 credits for 5 s at 720p, 189 at 1080p); Kling O1 Reference takes up to 7 references (73 credits for 5 s); Kling v3 Standard animates one start frame (86 credits for 5 s with sound).
- Ownership: the US Copyright Office says AI output is protected "only where a human author has determined sufficient expressive elements", "not the mere provision of prompts". Redraw or edit the final design yourself and start using it as a trademark.
Why AI mascots drift, and what fixes it
A video or image model doesn't remember your character between runs. It reads the prompt and whatever images you attach, and fills every gap with its own guess. "A friendly orange fox in a blue scarf" leaves the snout length, the eye shape, the scarf knot, the tail and the proportions open, so every generation fills them differently. That is the drift people notice by the third post.
There are two fixes, and you want both:
- Fewer open choices. A mascot with a simple silhouette, two or three colours and one signature detail has less to drift. A fox with a three-colour gradient, freckles and a patterned scarf has much more.
- An image, not words, as the source of truth. Once one design is approved, attach it to every generation. Words describe what changes (pose, scene, action); the reference carries what must not change.
The deeper mechanics, and where each model breaks, are in what is character consistency in AI.
The workflow on an 8frame board
Step 1: write the brief as constraints
Before any generation, write down what may not change. One short block you will paste into every prompt:
- species or type, body shape, head-to-body ratio
- exactly two or three colours (hex values if you have a brand palette)
- one signature item (scarf, cap, badge) and where it sits
- line style: flat vector, soft 3D, hand-drawn, plush toy
- personality in three words, because it drives poses and expressions
Keep the brief as a sticky note on the board next to the image nodes, so everyone who works on the mascot sees the same rules.
Step 2: explore wide, then pick one
Run 20 to 40 cheap candidates before you commit. Seedream 5 Lite at 6 credits an image and Nano Banana 2 at 8 credits for a 0.5K draft are the low-cost tools for this stage. Put the candidates side by side on the board, mark favourites with comments, and pick one. Don't average two designs in your head: choose one image as the parent.
Prompt:
Brand mascot concept: a small round fox, head as wide as its body, orange fur with a cream belly, a navy knitted scarf tied on the left, big oval eyes, short legs. Flat vector style, thick clean outlines, no gradients, plain white background, full body, front view, friendly and curious.
Step 3: turn the chosen design into a reference sheet
Attach the chosen image as a reference and ask for the same character from several angles on one canvas. ShotGrid makes a 3x3 contact sheet (nine frames in one image) at 4K for 42 credits and takes 1 to 10 reference images, which suits an expression sheet. For a clean turnaround (front, three-quarter, side, back), an image model with the parent attached works too.
Prompt:
Character turnaround sheet of the fox in the reference image: front, three-quarter, side and back views in a row, same proportions, same colours, same navy scarf tied on the left in every view. Flat vector, thick outlines, plain white background, even spacing, no text.
Check the sheet against the brief line by line: scarf side, ear shape, colour count, belly patch. Regenerate any view that breaks a rule, and keep the rest. This sheet is now the master. Remove the background (12 credits an image) if you want cut-out views for layouts.
Step 4: test the mascot in real situations
Before video, make six to ten stills of the mascot doing what it will do for the brand: holding the product, waving from a corner of a banner, reacting with surprise, sitting on a shelf. Attach the master sheet every time. Which model to pick depends on the job:
- Seedream 5 Lite takes the most reference images on 8frame (up to 14), so you can attach the whole sheet plus the product photo and a style frame.
- Nano Banana 2 (up to 10 references) is the default Nano Banana model and is priced by size: 8, 11, 17 or 22 credits at 0.5K, 1K, 2K or 4K.
- GPT Image 2 (1 to 10 references) is the option when the image needs readable text, such as a sign the mascot holds; at its default high quality it is 29 credits for a square image and 20 for other ratios.
Prompt:
The fox from the reference sheet stands on a wooden counter next to the coffee bag from the product photo, pointing at the label with one paw, smiling. Keep the exact proportions, colours and the navy scarf on the left. Flat vector style, warm cream background, square frame, space at the top for a headline.
Step 5: animate from the sheet, not from text
Now the video models. Each one handles references differently, and this is where the prices and the catches matter:
| Model on 8frame | What it takes | Credits | The catch |
|---|---|---|---|
| Happy Horse (reference-to-video) | 1 to 9 character references plus a prompt | 95 for 5 s at 720p, 189 at 1080p | Opens at 1080p, so switch to 720p for drafts. It has had little use on 8frame so far, so we quote its price, not its results |
| Kling O1 Reference | up to 7 reference images | 73 for 5 s | Check the mascot's details in every take |
| Kling v3 Standard | one start frame (image-to-video only) | 86 for 5 s with sound, 57 silent | Holds what is in the first frame; angles it never saw are invented |
| Kling O1 Image | start frame, and an optional end frame | 73 for 5 s | Useful when the clip must end on an exact pose |
A practical split: use a start-frame model (Kling v3 Standard or O1 Image) when the shot begins on one of your approved stills, and a reference model (Happy Horse or O1 Reference) when the mascot has to appear in a scene that doesn't exist yet. The two models side by side are compared in Happy Horse vs Kling v3 for reference-to-video.
Prompt:
The fox from the reference images walks into frame from the left across a sunny kitchen counter, stops next to a coffee cup, looks at the camera and waves. Same proportions, orange fur, cream belly, navy scarf on the left. Flat 2D animation style, static camera, 9:16.
Join clips with the Montage node (2 to 10 clips, 10 credits a run), but note the merged file has no sound: add music or a voiceover on the Video Editor timeline afterwards.
Step 6: keep a mascot library
Every approved output goes back on the board in a "canon" frame: the sheet, the best stills, the best clips. When someone on the team makes a new piece, they attach images from that frame, not from older drafts. Inviting teammates to the board needs a Creator plan (up to 3 members) or Agency; credits are charged to whoever runs the generation. For a longer campaign with several characters, see how to keep characters consistent across a campaign.
Do you own an AI-made mascot?
This is general information, not legal advice; ask a lawyer before you put serious money behind a character.
Copyright (US). In Part 2 of its report on AI (January 2025), the US Copyright Office concluded that AI outputs can be protected "only where a human author has determined sufficient expressive elements". That can be the case when a human-authored work is perceptible in the output or "a human makes creative arrangements or modifications of the output, but not the mere provision of prompts". In practice: a mascot that exists only as a prompt result has weak copyright footing. A designer redrawing the approved design as a vector file, or making real creative changes to it, puts human authorship into the version you actually use.
Trademark (US). The USPTO describes a trademark as "any word, phrase, symbol, design, or a combination of these things that identifies your goods or services", and says you become a trademark owner as soon as you start using it with your goods or services, with registration adding nationwide rights. A mascot used consistently on packaging, posts and ads is the kind of design that can work as a mark, which is one more reason to keep it consistent.
Don't imitate. Never prompt with the name of an existing mascot or character, or a description close enough to read as one. Besides the legal risk, a mascot that reminds people of someone else's doesn't build your brand.
Disclosure. Platforms have their own rules for AI-generated media in ads; see do you have to disclose AI-generated ads.
Common mistakes
- Changing the design in the prompt. "Same fox, but cuter" quietly creates a second mascot. Edit the master sheet on purpose, or don't change it.
- Too many references of different versions. Attaching an old draft next to the master tells the model both are right. Attach only canon images.
- Starting with video. Clips cost more than stills; find the problems at 6 to 11 credits an image, not at 86 to 189 a clip.
- Detailed patterns. Stripes, logos on clothing and small text drift first. Keep the signature item simple.
FAQ
What is the best AI tool to create a brand mascot?
The model matters less than the method: design once, then generate everything from an approved reference image. Choose image models by how many references they take (Seedream 5 Lite up to 14, Nano Banana 2 and GPT Image 2 up to 10 on 8frame) and video models that accept character references, such as Happy Horse (1 to 9) or Kling O1 Reference (up to 7).
How do I keep an AI mascot consistent in videos?
Start each clip from an approved still (Kling v3 Standard, 86 credits for 5 s with sound) or attach the reference sheet to a reference-to-video model (Happy Horse, 95 credits for 5 s at 720p). Repeat the fixed traits in the prompt, describe only the action and the scene, and check every take against the sheet before you use it. More techniques are in the AI character consistency workflow.
Can I copyright a mascot made with AI?
In the US, the Copyright Office says prompts alone are not enough; protection depends on human expressive contribution, such as creative modifications or arrangement of the output. Many brands have a designer redraw or refine the approved AI design. Trademark rights come from using the mascot with your goods or services, per the USPTO.
Can an AI mascot talk?
Yes, with two steps: a voice track (ElevenLabs text to speech on 8frame, 13 credits per started 1,000 characters) and a lip-sync tool. The lip-sync tools on 8frame are built around a portrait and an audio track; we haven't tested them on cartoon characters, so try one short take before you plan a series around it.
How much does it cost to make a mascot with AI?
Per generation on 8frame: 6 credits for a Seedream 5 Lite image, 11 for Nano Banana 2 at 1K, 42 for a ShotGrid sheet, 73 to 189 for a 5-second clip depending on the video model and resolution. Plans start at $19 a month for 1,000 credits. How many generations you need depends on how many rounds of changes the design goes through.
Sources
- US Copyright Office, "Copyright Office Releases Part 2 of Artificial Intelligence Report" (January 29, 2025): copyright.gov/newsnet/2025/1060.html, and the list of reports and registration guidance at copyright.gov/ai, checked 2026-11-04
- USPTO, "What is a trademark?": uspto.gov/trademarks/basics/what-trademark, checked 2026-11-04
- 8frame prices, presets, defaults and limits (reference image limits, ShotGrid, Happy Horse, Kling, Montage, collaborators): the 8frame app, checked 2026-11-04
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.