Seedream 4.0 guide: editing, prompts & when to use 5.0
Seedream 4.0 is ByteDance's September 2025 unified image generation and editing model. It remains available in Magic Hour for prompt-based creation, controlled edits, multi-image references and output up to 4K. For a new final-production evaluation, also test Seedream 5 Pro, released July 8, 2026, especially when dense layouts, multilingual text or annotation-guided editing matter. Keep 4.0 when its lower cost, 4K route or known behavior fits the job better.
Quick answer: Upload the clearest source and only the references that have a defined role. Describe the exact change, say what must remain unchanged, generate at draft resolution, then inspect identity, text, product details, geometry and edit boundaries at full size. Version names are not interchangeable: record the selected model, resolution, credits and accepted output.
Seedream versions: what to use now
Version | Official status | Use it for | Magic Hour route |
|---|---|---|---|
Released September 2025; earlier generation and editing route | Lower-cost known workflow, multi-reference work or 4K output | Seedream 4 model page | |
Later 4.x model with provider-claimed consistency and typography improvements | Existing 4.5 workflows or 4K editing in Magic Hour | Seedream 4.5 model page | |
Released February 13, 2026 with reasoning and optional search capabilities | Provider routes that expose knowledge-led or search-assisted creation | Not currently listed in Magic Hour's model catalog | |
Released July 8, 2026 as ByteDance's current professional model | Dense layouts, multilingual text, high-detail generation and precise edits | 1K and 2K generation or editing in Magic Hour |
ByteDance's current model catalog lists both Seedream 5.0 Lite and 5.0 Pro. Magic Hour currently lists Seedream 4, 4.5 and Seedream 5 Pro. Do not infer that a model is available in a product merely because ByteDance announced it. Check the exact picker or API schema before committing a workflow.
Seedream 4.0 facts at a glance
- **Core tasks:** text-to-image, prompt-based image editing, multi-image composition and multi-output generation.
- **Resolution:** ByteDance documents output up to 4K. Magic Hour offers 1K, 2K and 4K on its Seedream 4 model page; the API also documents 640px for eligible editor requests.
- **Reference limit:** ByteDance describes up to a dozen model-level references. Magic Hour's current editor API documents a maximum of nine additional images for `seedream-v4`.
- **Current Magic Hour model-page cost:** 25 credits at 1K, 40 at 2K and 50 at 4K. The editor API lists `seedream-v4` from 40 credits per image, so price browser and API jobs separately.
- **Best fit:** controlled image revisions, product compositions, posters, character references and variations that need stronger continuity than a one-line generation prompt.
These facts were checked on September 11, 2026. Model availability, plan access and credit pricing can change; use Magic Hour's Seedream 4 model page, pricing page and the generation estimate shown before submission for a current quote.
How to use Seedream 4.0 in Magic Hour
- Open the AI Image Editor and select Seedream 4 when that model is available in the picker.
- Upload the highest-quality source image and only the references that have a clear job. Crop out unrelated UI, borders and screenshots when possible.
- Write one instruction that names the subject, the requested change and the details that must stay fixed.
- Choose the aspect ratio and resolution required by the final placement. Use a lower-cost draft first when the composition is still uncertain.
- Review the full-size output. Check faces, hands, logos, exact text, product geometry, reflections and the boundary of every changed region before downloading or publishing.
If the first result misses the brief, change one part of the instruction at a time. A short correction such as “keep the bottle label and camera angle unchanged; replace only the background” is easier to diagnose than a new prompt that changes the subject, setting, lighting and style together.
A prompt structure that is easy to control
Use this order: **task + subject + requested change + details to preserve + composition + constraints**. Seedream does not require invented role weights or magic words. State each reference's purpose in ordinary language.
**Example:** “Edit the first image. Keep the person's face, hairstyle, pose and black jacket unchanged. Replace the office with the rainy Tokyo street in the second image. Match the street's blue evening light, keep a waist-up composition and do not add text or logos.”
A useful prompt identifies both the change and the acceptance criteria. “Make this better” gives the model little guidance. “Remove the glare from the glasses while preserving the eye shape, skin texture, crop and background” creates a result you can evaluate.
How to use reference images
Use the fewest references that fully define the result. One image can establish the subject. A second can establish the setting or composition. A third can show a product, garment or visual treatment. More inputs can help a complex composite, but they also create more opportunities for the model to blend the wrong detail.
- **Identity reference:** say which face, character or product must remain recognizable.
- **Composition reference:** name the framing, camera angle or object placement to follow.
- **Style reference:** identify the lighting, texture or medium to borrow without asking the model to copy protected brand assets.
- **Content reference:** say which object, background or garment should be inserted and where it belongs.
Do not assign made-up numerical weights such as “character 1.0” or assume that words such as “series” activate a hidden consistency mode. If consistency matters, repeat the details that must not change and retain the accepted output as the reference for the next edit.
Five copyable Seedream 4.0 editing prompts
1. Replace a product background
“Keep the product, label, cap, proportions and camera angle exactly as shown. Replace only the background with a warm limestone bathroom counter at sunrise. Add a soft contact shadow beneath the bottle. Do not change the packaging text, logo or product color.”
2. Create a consistent character variation
“Use the person in the first image as the identity reference. Preserve facial structure, skin tone, hairstyle and age. Change the outfit to the green jacket in the second image and place the subject in a chest-up studio portrait on a neutral gray background.”
3. Combine two images
“Place the chair from the first image in the room from the second image. Keep the chair's shape, material and color unchanged. Match the room's camera height, perspective and window light. Position the chair to the right of the coffee table with a realistic floor shadow.”
4. Edit text in a poster
“Keep the poster's layout, colors and illustration. Replace the headline with ‘SUMMER NIGHT MARKET’ and the date with ‘18 JULY’. Use the same hierarchy and centered alignment. Do not add other words.” Then inspect every character; generated typography can still contain spelling or spacing errors.
5. Make a restrained style change
“Preserve the subject, pose, crop and background objects. Change the rendering to a soft editorial gouache illustration with visible paper texture, muted earth colors and simple shadows. Keep the face recognizable and do not add captions, borders or signatures.”
What to check before using the result
- **Identity:** compare the face, product silhouette and distinctive details with the source.
- **Text and logos:** zoom in and verify spelling, spacing, trademark shape and placement.
- **Geometry:** inspect fingers, repeated objects, straight edges, reflections and perspective.
- **Edit boundary:** look for halos, seams, color shifts and duplicated textures around the changed area.
- **Claim accuracy:** do not use a generated image as evidence of a physical product's size, fit, safety or performance.
- **Rights:** confirm that you may use the source images, brands and likenesses in the intended context.
A 4K export increases pixel dimensions; it does not prove that small text, faces or product details are correct. Approve the composition at draft resolution, then review the final-resolution file again.
Seedream 4.0 versus Seedream 4.5
ByteDance describes Seedream 4.5 as improving reference-detail consistency, typography and multi-image editing. Those are provider claims, not an independent Magic Hour benchmark. Choose 4.5 when those improvements match your task and the model is available to your plan; keep 4.0 when an existing workflow, accepted output or known cost matters more than switching models. Test both with the same inputs and acceptance checklist before deciding.
Using Seedream 4.0 through the API
Magic Hour exposes Seedream 4 for image generation and editing through its API. The current AI Image Editor reference lists `seedream-v4`, supported resolutions and aspect ratios, plan access, credit cost and maximum additional images. Use the API schema as the source of truth for request fields; do not copy browser-interface assumptions into an integration.
For a model-specific entry point, review the Seedream 4 API page. Store the returned project ID, poll the documented status endpoint and retain inputs plus accepted outputs when you need repeatable production QA.
Seedream 4.0 FAQ
Can Seedream 4.0 edit an existing image?
Yes. ByteDance documents prompt-based single-image and multi-image editing. Upload the source, describe the specific change and name the details that must remain fixed.
How many reference images can I use?
ByteDance describes support for up to a dozen references at the model level. Magic Hour's current AI Image Editor API lists up to nine additional images for `seedream-v4`. A browser product or another provider may apply a different limit, so verify the surface you actually use.
How much does Seedream 4.0 cost in Magic Hour?
Magic Hour's Seedream 4 model page currently lists 25 credits for 1K, 40 for 2K and 50 for 4K image creation. The AI Image Editor API lists `seedream-v4` from 40 credits per image. Check the selected workflow's live quote because browser and API pricing are separate.
Can Seedream 4.0 generate accurate text?
It can generate and edit text, but accuracy is not guaranteed. Keep the requested wording short, quote the exact copy in the prompt and inspect every character before publication. Use a design editor for legally sensitive or pixel-exact typography.
Can I use Seedream 4.0 images commercially?
Magic Hour allows commercial use on paid plans, subject to its current terms and your rights to the inputs and depicted people or brands. Model access does not give you rights to third-party source material.
Primary sources
- ByteDance Seedream 4.0 product page
- ByteDance Seedream 4.0 release note
- ByteDance Seedream 4.5 product page
- Magic Hour Seedream 4 model page
- Magic Hour AI Image Editor API reference
Finish your Seedream 4.0 edits in Magic Hour
Upload an image and describe the specific edit you want. Review the result before downloading. Free usage and model access vary; commercial use requires a paid subscription.
Try AI Image Editor for FREEFor a side-by-side model comparison, see the best AI image editing models for reference images.
Aastha Kochar has spent 5+ years creating content for B2B and B2C SaaS brands in the AI and MarTech space. She is well-versed with AI-powered content tools and offers deep comparisons after trying and testing every tool.
Her work has helped companies increase organic traffic, earn AI citations, and most importantly — turn readers into users.
With a bachelor's and master's degree in Journalism and Mass Communication, she brings strong research skills, authentic storytelling, and a deep understanding of what makes audiences actually care about what they're reading.
