What’s new
- Reasons before it draws: ByteDance says Lite adds deep-thinking capability, working out what you mean before generating instead of pattern-matching on keywords.
- Web-connected retrieval: the first Seedream model with real-time web search, so time-sensitive content can pull in current information before generating.
- Handles complex instructions: multi-step prompts, multiple reference images, local-edit instructions and batch style requests are all supported.
- Up to 15 images per request: the n parameter goes from 1 to 15 (reference images plus n together must stay at or under 15), far more flexible than Pro.
What it’s good for
- Time-sensitive content that needs current information pulled in before generating.
- Information visualization for office and education use: charts, diagrams, complex layouts.
- Batch generation when you need several style variants or a set of related assets at once.
Prompting tips
- Vague instructions are fine: ByteDance says it can work with loosely worded requests, no need to pile on technical jargon.
- Break multi-step reasoning into steps: for scenes that require "figuring something out" (like a chess position or an assembly diagram), spell out the logic step by step.
- Say what each reference image is for: note whether it's a style reference or a color reference so the model applies it correctly.
- Ask directly when you need current info: name what should be looked up (weather, a trending topic) right in the prompt.
- Use n for variant sets: generating several options in one request beats calling it repeatedly.
A four-step diagram showing "how to assemble a bookshelf", laid out left to right,
each step numbered with a short caption, clean line-art style, light gray background:
Step 1 place the side panels; Step 2 fit the shelves; Step 3 attach the back panel; Step 4 secure the hardware.
Known limits
- ByteDance itself calls this "a relatively small model," with room to improve on structural stability, realism and aesthetics.
- Reference image count and output count share one cap (combined at most 15), so more references leaves room for fewer outputs.
- The 9:21 portrait ratio isn't supported (21:9 landscape is).
Using it in Nomi
On the generation canvas, add a image card, pick “Seedream 5.0 Lite” 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.
