What’s new
- Dense layouts in one pass: ByteDance says it produces near-ready professional layouts for infographics, posters and UI/product mockups packed with both data and text.
- Pixel-level interactive editing: point selection, lasso, sketch coloring, recoloring, material swaps, layer separation and multi-image fusion — fix one part instead of regenerating the whole image.
- More realistic materials and lighting: better skin and material rendering, plus support for photography-style camera moves.
- A dozen-plus languages, natively: French, German, Russian, Japanese, Korean, Spanish, Arabic and more can be typed directly, no translation to English needed first.
What it’s good for
- Text-and-data-heavy posters, infographics and UI/product mockups that need to come out already laid out.
- Commercial retouching where you keep returning to fix one detail: a color, a material, one isolated layer.
- Localized marketing assets for global markets, generated directly in the target language.
Prompting tips
- Cover five elements: subject, action/pose, environment, style, and technical details (lighting/camera/resolution) — the structure Dreamina's own guide recommends.
- Put in-image text in quotes: e.g. text reading "Good Morning Café" renders more reliably than unquoted text.
- Name the region when editing: with point-select or lasso edits, say exactly where and what should change.
- Spell out materials and lighting: not "nice light" — say what material, what light source, from which direction.
- Tag each reference image: note what person or style each of up to 10 reference images is for.
- Expect one image per run: this tier doesn't support n>1, so generate variations one request at a time.
Promotional poster for a boutique coffee shop, warm vintage tones. Headline text reads
"Morning Brew Café · Grand Opening", subhead "Pour-over · Roastery · Slow mornings".
A steaming pour-over coffee on a wood table, soft morning side light, shallow depth of field,
empty space at the bottom for an info panel. 16:9.
Known limits
- Only one image per request — no n>1, no sequential batches, no streaming.
- Up to 10 reference/input images; more than that is rejected.
- ByteDance says fine-grained text rendering and pixel-level editing consistency still have room to improve.
Using it in Nomi
On the generation canvas, add a image card, pick “Seedream 5.0 Pro” in the model picker and choose a mode (Text to image, Image editing). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
