Write the exact words you want rendered, label all six references, and change one thing at a time.
- Order: Subject → Style & mood → Composition → Key details → Reference.
- For any text in the image, write the exact words. Do not describe them.
- Up to six reference images — and label what each one contributes.
- No rigid formula published, unlike Seedance. The guidance is to keep it "clear, structured and focused".
- 2K for digital, 4K for print.
This sheet quotes the vendor's own published prompting guidance.
The formula
Template:
[subject, stated explicitly] [style and mood] [how elements sit in the frame] [lighting, colour, texture, must-have items] [+ references, each labelled with its job]
Worked example:
A ceramicist centring clay; documentary realism, quiet and warm; subject left-of-centre with the window behind; north light, muted earth palette, visible clay texture; poster text reads exactly "OPEN STUDIO — SATURDAY"
Breakdown — what each part does
Subject
Person, product, object or scene — stated, never implied. ByteDance name an ambiguous subject as a specific mistake.
Visual style & mood
One direction: cinematic, realistic, minimal, futuristic, editorial, artistic. Not several at once.
Composition / layout
Where things sit in the frame. This model's positioning logic is a strength worth directing.
Key details
Lighting, colours, environment, textures, must-have items — the supporting specifics.
Reference images
Up to six. Say what each contributes: colour scheme, layout, pose, style.
Does it take a negative prompt?
ByteDance's stated approach is positive and diagnostic: describe what you want, and when something goes wrong, fix that one thing rather than pre-loading exclusions. Their explicit warning is against conflicting instructions in a single prompt — which is what a long exclusion list becomes.
What not to do
Released 10 February 2026 with materially better text: small fonts render more accurately, character repetition is reduced, and the bias toward bold is corrected. This is the model to reach for when the image contains words.
Editing is by partial selection and pen input rather than layers — you can alter one region without disturbing the rest of the frame.
The checklist
Before you send it:
Dreamina (ByteDance), Seedream 5.0 Pro guides, dreamina.capcut.com. Checked 25 Aug 2026.
Model versions and vendor documentation both move. Re-read the source before relying on a specific number. Errors are logged at corrections.
Every other sheet in this set: all model cheat sheets.