Seedream 5 Prompting Guide
Prompt Seedream 5 as a design brief: deliverable, product lock, quoted type, one-change edits. Run Pro 6cr / Lite 3cr on Versely.
Seedream 5 prompting is not a mood board caption. Treat the box like a design brief: deliverable, subject, composition, materials, quoted on-image copy, and an explicit preserve list when you edit. On Versely you run four ByteDance rows from the Seedream 5.0 Pro family: Pro and Pro Edit at 6 credits, Lite and Lite Edit at 3 credits, then iterate the brief instead of restacking adjectives.
This guide is the general Seedream 5 / Seedream 5 Pro technique hub. For type-heavy posters and hierarchy language see the deep-dive sibling Seedream 5 prompting for typography and layout. For dense data boards see Seedream 5 Pro dense infographics. Stills siblings that use the same brief discipline: GPT Image 2 prompting guide and Nano Banana 2 prompting guide.
The design brief stack
Only deliverable and subject are mandatory. Add the other slots in the order your last still failed.
- Deliverable - product still, packaging comp, poster, UI mock, lifestyle plate, edit pass
- Subject - who or what must stay readable and on-SKU
- Scene - place, materials, light direction, color temperature
- Composition - framing, viewpoint, placement, aspect ratio, negative space
- Text - quoted strings, hierarchy, placement (or no text)
- References - each Image N gets one narrow job
- Style - photoreal product, editorial, print, CG; keep it short
- Constraints - change-only scope plus preserve list (exclusions live in the prompt)
Write it as labeled lines if that helps you edit. Seedream 5 reads natural-language briefs well. What it does not forgive is five conflicting ads glued with commas, or an edit that never says what must stay locked.
Pro, Lite, and Edit: pick the row first
| Row | Credits | Use when |
|---|---|---|
| Seedream 5.0 Pro | 6 | Hero stills, packaging, multilingual type, dense layout |
| Seedream 5.0 Lite | 3 | Fast direction tests, batch variants, cheap comps |
| Seedream 5.0 Pro Edit | 6 | One-region fixes on an approved master |
| Seedream 5.0 Lite Edit | 3 | Cheap edit drafts before you spend Pro Edit |
Draft on Lite when you are still hunting composition. Lock the winner, then move to Pro for type, label fidelity, and final materials. Prefer Edit rows when the master is already right and only one variable should move. Catalog details stay on /models/seedream-5-pro; run everything from /tools/text-to-image.
Product stills and packaging comps
Name the SKU geometry before you name the vibe. Call out finish (matte, foil, glass, condensation), camera height, and what must not invent itself on the label.
Weak:
A beautiful premium soda can looking luxurious and viral.
Stronger:
DELIVERABLE: 4:5 ecommerce hero for Instagram.
SUBJECT: Tall aluminum citrus soda can, frosted finish, pull-tab intact.
SCENE: Wet slate surface, soft daylight from upper left, subtle condensation.
COMPOSITION: Three-quarter view, can in the lower-right third, empty upper-left for later type.
STYLE: Photorealistic commercial product photography, sharp focus on the label edge.
CONSTRAINTS: No extra cans, no watermark, no invented logos, no people.
When you attach a pack shot, uploading is not the instruction. Label the asset and give it one job.
REFERENCE: Image 1 is the bottle to place. Preserve shape, cap, label art, proportions, and exact color. Do not copy Image 1's white seamless backdrop or lighting.
SCENE: Marble bathroom counter, folded linen, one eucalyptus sprig. Soft window light from the left.
COMPOSITION: Bottle left of center, three-quarter view.
CONSTRAINTS: No rewritten label text, no extra bottles, no watermark.
Worked prompt 1 (packaging still with short type)
DELIVERABLE: 1:1 packaging hero, soft gray seamless.
SUBJECT: Matte navy coffee bag with a cream valve, front-facing, slight three-quarter tilt.
TEXT: Large condensed cream headline on the bag reads "ALTITUDE". Smaller line under it reads "SINGLE ORIGIN". Two lines only. Quote case as written.
LIGHT: Softbox from upper left, gentle fill from the right, realistic fabric grain.
CONSTRAINTS: No other text, no barcode invent, no watermark, no props that hide the lockup.
For long ingredient blocks or legal microcopy, generate the bag and composite body type externally. Short brand lines belong in the still; paragraphs usually do not.
Edit prompts: change only X, preserve the rest
Generation prompts and edit prompts need different shapes. An edit that only names the change invites a full redesign. State the target, the change, and a preserve list every time.
Weak: Make the can red.
Stronger:
EDIT: Change only the can body from matte black to brushed silver.
TEXT: Replace the central flavor label with "YUZU LIME" in bold condensed sans.
PRESERVE: Exact can shape, size, position, condensation, highlights, shadows, hand placement, background, camera angle, and existing logo.
CONSTRAINTS: Do not alter composition. Do not add objects. No watermark.
One variable per pass beats five stacked wishes. Restate the preserve list on every follow-up. When the UI exposes selection, box, or lasso tools, use them to name the region, then keep the same change-plus-preserve language in text.
Worked prompt 2 (localized campaign edit)
EDIT: Using Image 1 as the approved poster master.
CHANGE: Replace only the English headline "CREATE WITHOUT LIMITS" with the Spanish headline "CREA SIN LIMITES".
PRESERVE: Product, person, pose, background, lighting, logo, palette, spacing, and all other text.
CONSTRAINTS: Keep the same headline zone and hierarchy. Natural line breaks for Spanish. No other redesign.
Proofread every localized string with a fluent speaker before you ship. Native multilingual rendering is strong; brand and legal review still sit with you.
Layout and type without rewriting the typography post
Seedream 5 is strong at readable short copy when you quote it, name hierarchy, and give type a quiet zone. Keep the habits brief here; the full type dialect lives in Seedream 5 prompting for typography.
- Put required wording in straight quotes.
- Label roles: headline, subhead, caption, CTA.
- Place each string in a zone (
top third,cream band across the bottom). - Prefer typeface classes (
condensed sans,editorial serif) over commercial font names. - Keep load-bearing copy short; break long lines into separate quoted elements.
Worked prompt 3 (poster layout)
DELIVERABLE: 9:16 launch poster, warm editorial.
SCENE: Ceramic mug on linen, soft window light in the center zone.
TEXT: Top third large high-contrast serif headline "SLOW MORNINGS". Bottom cream band medium sans "New single-origin roast" with small caption "Launching March 3".
COMPOSITION: Generous margins. Type on quiet bands, not on busy fabric grain.
PALETTE: Muted terracotta and cream.
CONSTRAINTS: No other text, no watermark, no QR codes.
If the job is mostly lettering, stop here and open the typography sibling. If the job is mostly SKU lock and scene, stay on this hub.
Natural language, use case, and visual markers
Prefer full sentences over keyword stacks. Naming the job helps: Design a 4:5 ecommerce hero for a citrus soda beats a bag of mood words. For hard-to-describe transforms, example-based editing works: show a before/after pair, then ask the model to apply the same operation to a third image. When the composer allows doodles, arrows, or colored regions on the input, reference those markers in the prompt so spatial edits stay deterministic.
Keep style to three to six concrete terms after the subject. Photoreal product language, lens feel, and light direction stick better than empty intensifiers like "premium" or "viral."
Failure modes
| Failure | Likely cause | Fix in the brief |
|---|---|---|
| Product redesigned mid-edit | No preserve list | Change only X; restate shape, label, camera, lighting |
| Rewritten or gibberish type | Unquoted copy, busy background | Quote strings; quiet band; shorten lines |
| Generic centered catalog look | No composition or aspect | Name framing, placement, ratio, negative space |
| References fight each other | Multiple images, no roles | One job per Image N; say what not to copy |
| Edit becomes a new generate | Too many variables at once | One change per pass; lock the master |
| Soft edges after approval | Asking the model to "fix" cleanup | On-device /free-tools/image-upscaler or /free-tools/background-remover |
| Watermark or fake UI chrome | Web-trained leak | Constraints: no text, no watermark, no logos |
Exclusions live in the prompt text. Targeted absences beat long boilerplate stacks.
Skip the dialect: brief the Versely agent
You can learn this stack. On Versely you can also skip memorizing every Seedream dialect and brief the agent instead of writing prompts: content type, photos or references, the edits you will accept, target platforms, and a budget ceiling. Name the row when it matters ("use Seedream 5.0 Pro" or "Seedream 5.0 Lite Edit"). The agent plans the job; you approve the plan.
A prompt still wins when you are hand-tuning one hero still. A brief wins when the job is bigger than one generate: variants, localization, cleanup, and a post path with a spend cap.
After the take: edit, post, collections
When the Seedream pass is close enough:
- Finish cutouts and sharpening on-device with /free-tools/background-remover and /free-tools/image-upscaler when those are the remaining gaps. Use /free-tools/object-eraser for stray props.
- Prefer Pro Edit or Lite Edit for one-region copy or material fixes on an approved master instead of regenerating the whole frame.
- Optional motion: send a winning still into /tools/ai-video-generator as image-to-video. Prompt the motion brief on the video row; do not ask Seedream to invent a timeline.
- Upload or schedule to the social accounts Versely already connects for your workspace. If a network is not connected, export the PNG or JPEG and post from the native app.
- Save keepers into a Versely collection so the next brief reuses the same product refs, winning prompt lines, and approved masters.
Honest limit: Seedream 5 is a stills and edit model, not a full desktop publishing system and not a video timeline. Board separate generates for separate setups. Credits apply on every generate.
FAQ
What is the best Seedream 5 prompt formula?
Deliverable and subject first, then scene, composition, quoted text, reference roles, style, and constraints (plus a preserve list on every edit). Drop any slot you do not need. Add slots in the order your last failure showed up.
When should I use Pro versus Lite?
Use Lite (3 credits) to hunt composition and batch cheap variants. Use Pro (6 credits) when label fidelity, multilingual type, dense layout, or final materials matter. Move approved masters to Pro Edit or Lite Edit for one-change revisions.
How do I edit without redesigning the whole image?
Name the exact change, then list everything that must stay fixed: shape, pose, lighting, logo, camera, other text. One variable per pass. Restate the preserve list on follow-ups.
How do I get readable text in Seedream 5?
Quote the exact string, name hierarchy and placement, keep lines short, and give type a quiet zone. For the full typography dialect see Seedream 5 prompting for typography. Proofread before you ship.
Where do I run this on Versely?
Open the text-to-image tool with a Seedream 5 row selected, or start from the Seedream 5.0 Pro model page. Animate approved winners later in the AI video generator via image-to-video.
How is this different from GPT Image 2 or Nano Banana 2 prompting?
Same stills-brief discipline, different strengths. Seedream leans packaging, multilingual type, dense layout, and controlled edits. Compare technique notes with the GPT Image 2 prompting guide and Nano Banana 2 prompting guide when you are routing the same brief across rows.
Takeaway
Prompt Seedream 5 like a short design brief, not like a vibe caption. Lock deliverable and composition, quote every load-bearing string, give each reference one job, and write change-plus-preserve on every edit. Draft on Lite, finish on Pro, revise on Edit. Run the rows on Versely, or hand the agent a job brief when you would rather not memorize the dialect. Clean winners on-device when upscale or remove-bg is enough, animate keepers via image-to-video when motion is next, and file keepers in a collection so the next still starts warmer.