A quick look at what Seedream 4.5 produces out of the box — no upscaling, no post-production.
Serif title, tight negative space, three-line small-caps event copy — rendered in a single pass.
Dark-green-on-green palette with calm, copy-heavy hierarchy for brand storytelling.
Photographic subject + hand-drawn line illustration, three compositions in one image.
Layered typography, motion blur, and geometric burst effects anchored on the running silhouette.
16:9 promo image for a SaaS product — clean information hierarchy and lifestyle framing.
Gradient background with ingredient lockup — small-text precision in a marketing hero.
Consistency with Reference Image: Preserves facial features, lighting, color tone, and fine details when editing reference images.
Accurate Multi-image Editing: Composes up to ten input images in a single prompt while keeping each subject identifiable and consistent.
Poster Layout & Logo Design: Designer-level composition, hierarchy, and spacing — from editorial posters to e-commerce KVs.
Dense & Stylized Text Rendering: Small captions, dense labels, handwritten fonts, and in-image language swaps all stay legible.
Material Transformation & Localized Edits: Swap fabrics, materials, or colors on a specific region while leaving everything else untouched.
Seedream 4.5 enforces reference fidelity at every stage of editing. Faces retain identity across changes in pose, material, and background. Lighting moods carry through style transfers. Even when you swap materials or recolor localized regions, the unedited parts of the image remain untouched — giving you professional retouching behavior without a round-trip through external tools.
Example 1
Subject Reference
Prompt
Only keep the character in the green outline, and delete the other characters.
Result
Example 2
Subject Reference
Prompt
3D render, miniature world. Night scene, dark blue sky, moonlight on the grass. Warm orange light glowing from the windows of the yurts.
Result
Subject Reference
Prompt
Result
Only keep the character in the green outline, and delete the other characters.
3D render, miniature world. Night scene, dark blue sky, moonlight on the grass. Warm orange light glowing from the windows of the yurts.
Instead of generating from scratch, Seedream 4.5 reads every reference you upload, decides which subject or element to pull from which image, and assembles them under a single natural-language instruction — replace the product in Figure 1 with the product in Figure 2; copy the title from Figure 3 to the top of the frame; and so on. Subject identification stays stable across inputs, so swaps and composites don't drift into the wrong character, product, or brand.
Example 1
Subject Reference
Prompt
Replace the product in Figure 1 with that in Figure 2. For the title, copy the text in Figure 3 to the top of the screen; the title should have clear contrast with the background but not be overly eye-catching.
Result
Subject Reference
Prompt
Result
Replace the product in Figure 1 with that in Figure 2. For the title, copy the text in Figure 3 to the top of the screen; the title should have clear contrast with the background but not be overly eye-catching.
Seedream 4.5 handles full-stack layout decisions that previous image models avoided: grid alignment, leading, type hierarchy, negative space, foil and emboss treatments, and the interaction between illustrative subject and typographic block. Works with serif, sans, display, decorative, handwritten, and multi-language type.
Example 1
Prompt
Minimalist art exhibition poster with an off-white background. A black line drawing of a human portrait is positioned slightly to the right of the center. The title "A/W ART FAIR" is written in serif font at the top left corner. Three lines of extremely small font at the bottom left corner indicate the exhibition period, venue, and exhibition story, aligned to the left with even spacing.
Result
Example 2
Prompt
A European vintage-style wedding invitation design. The card has a dark green base, with textured white paper embossed with delicate patterns layered on top—its edges adorned with elaborate lace-like carvings. The card features elegant gold foil typography with wedding details: the names of the bride and groom, date, venue, blessings, and invitation message. At the bottom of the card, a dark green dome-shaped European church paper cutout serves as decoration. Surrounding it are envelopes with botanical patterns and green flowers.
Result
Prompt
Result
Minimalist art exhibition poster with an off-white background. A black line drawing of a human portrait is positioned slightly to the right of the center. The title "A/W ART FAIR" is written in serif font at the top left corner. Three lines of extremely small font at the bottom left corner indicate the exhibition period, venue, and exhibition story, aligned to the left with even spacing.
A European vintage-style wedding invitation design. The card has a dark green base, with textured white paper embossed with delicate patterns layered on top—its edges adorned with elaborate lace-like carvings. The card features elegant gold foil typography with wedding details: the names of the bride and groom, date, venue, blessings, and invitation message. At the bottom of the card, a dark green dome-shaped European church paper cutout serves as decoration. Surrounding it are envelopes with botanical patterns and green flowers.
Text is rendered as typography, not pixels — small ingredient lists stay readable, handwritten Chinese preserves stroke flow, and in-image language swaps keep the surrounding layout intact. Handles mixed-script layouts (CJK + Latin) without garbling either side.
Example 1
Subject Reference
Prompt
Translate the image into Chinese, using a handwritten font, with all other content unchanged.
Result
Example 2
Subject Reference
Prompt
Beauty new product KV design with a pink-white gradient background. The product features soft light reflection. Large text at the top: Revitalize Skin Foundation Multi-Effect Repair. Small text for ingredient description on the right, arranged in sections: Sodium Hyaluronate 3% moisturizing factor; Niacinamide 2% brightening; Ceramides 1.5% repairing; Centella Asiatica + Calendula Extract 2.8% complex.
Result
Subject Reference
Prompt
Result
Translate the image into Chinese, using a handwritten font, with all other content unchanged.
Beauty new product KV design with a pink-white gradient background. The product features soft light reflection. Large text at the top: Revitalize Skin Foundation Multi-Effect Repair. Small text for ingredient description on the right, arranged in sections: Sodium Hyaluronate 3% moisturizing factor; Niacinamide 2% brightening; Ceramides 1.5% repairing; Centella Asiatica + Calendula Extract 2.8% complex.
Mask an area with a highlight, draw an outline, or just describe the region — Seedream 4.5 scopes the edit precisely. Lighting shifts from reflection to refraction when materials change; recolored words stay italicized where needed; unedited pixels stay bit-identical.
Example 1
Subject Reference
Prompt
Keep the model's pose and the flowing shape of the liquid dress unchanged. Change the clothing material from silver metal to completely transparent clear water (or glass). Through the liquid water, the model's skin details are visible. Lighting changes from reflection to refraction.
Result
Example 2
Subject Reference
Prompt
Change the word in the red highlighted area to blue italic text.
Result
Subject Reference
Prompt
Result
Keep the model's pose and the flowing shape of the liquid dress unchanged. Change the clothing material from silver metal to completely transparent clear water (or glass). Through the liquid water, the model's skin details are visible. Lighting changes from reflection to refraction.
Change the word in the red highlighted area to blue italic text.
Up to ten input images can be referenced in a single instruction — Seedream 4.5 picks the right subject from each and composes them consistently.
Example 1
Input images
Inputs (Figure 1–3)
Instruction
Replace the product in Figure 1 with that in Figure 2. For the title, copy the text in Figure 3 to the top of the screen; the title should have clear contrast with the background but not be overly eye-catching.
Output
Inputs (Figure 1–3)
Replace the product in Figure 1 with that in Figure 2. For the title, copy the text in Figure 3 to the top of the screen; the title should have clear contrast with the background but not be overly eye-catching.
Even when the clothing material, background, or embedded text changes, faces, lighting, and layout stay locked to the reference.
Example 1
Reference image
Instruction
Keep the model's pose and the flowing shape of the liquid dress unchanged. Change the clothing material from silver metal to completely transparent clear water (or glass). Through the liquid water, the model's skin details are visible. Lighting changes from reflection to refraction.
Result
Example 2
Reference image
Instruction
Generate a best-selling product display image in a modern e-commerce platform style. Top title: "Today's Hot Sale"; main product in the middle of the image; price "$9.99" below it. Overall background: a vertical gradient transitioning from orange-pink to light purple-blue.
Result
Example 3
Reference image
Instruction
Translate the image into Chinese, using a handwritten font, with all other content unchanged.
Result
Keep the model's pose and the flowing shape of the liquid dress unchanged. Change the clothing material from silver metal to completely transparent clear water (or glass). Through the liquid water, the model's skin details are visible. Lighting changes from reflection to refraction.
Generate a best-selling product display image in a modern e-commerce platform style. Top title: "Today's Hot Sale"; main product in the middle of the image; price "$9.99" below it. Overall background: a vertical gradient transitioning from orange-pink to light purple-blue.
Translate the image into Chinese, using a handwritten font, with all other content unchanged.
Every image below was generated or edited in a single pass from a natural-language prompt. No upscaling, no retouching.
Minimalist art exhibition poster with an off-white background. A black line drawing of a human portrait is positioned slightly to the right of the center. The title "A/W ART FAIR" is written in serif font at the top left corner. Three lines of extremely small font at the bottom left corner indicate the exhibition period, venue, and exhibition story, aligned to the left with even spacing.
A European vintage-style wedding invitation design. The card has a dark green base, with textured white paper embossed with delicate patterns layered on top—its edges adorned with elaborate lace-like carvings. The card features elegant gold foil typography with wedding details: the names of the bride and groom, date, venue, blessings, and invitation message. At the bottom of the card, a dark green dome-shaped European church paper cutout serves as decoration. Surrounding it are envelopes with botanical patterns and green flowers.
Black-orange dynamic style gym promotional key visual with layered typography elements. At the center stands an athlete's running silhouette, dynamic motion blur effect on limbs, and a concentrated posture that amplifies the sense of speed and power. The background features abstract geometric patterns in black, orange, and dark gray, with diagonal light streaks and energy burst effects to reinforce the dynamic vibe.
3D render, miniature world. Night scene, dark blue sky, moonlight on the grass. Warm orange light glowing from the windows of the yurts.
Keep the model's pose and the flowing shape of the liquid dress unchanged. Change the clothing material from silver metal to completely transparent clear water (or glass). Through the liquid water, the model's skin details are visible. Lighting changes from reflection to refraction.
Generate a best-selling product display image in a modern e-commerce platform style. Top title: "Today's Hot Sale"; main product in the middle of the image; price "$9.99" below it. Overall background: a vertical gradient transitioning from orange-pink to light purple-blue.
Three sets of minimalist creative advertising combinations: a fresh blueberry paired with hand-drawn black ink graffiti, featuring brisk and vivid lines; a fluffy handmade marshmallow paired with hand-drawn black ink graffiti, featuring soft wavy lines; a real ginkgo leaf paired with hand-drawn black ink graffiti, featuring elegant and smooth lines.
High-end fragrance product detail page, mainly in dark green and light green tones, with concise and elegant text and copy, conveying the tranquil atmosphere of mountain forests and open wilderness.
Based on ByteDance's internal MagicBench multi-dimensional evaluation — Seedream 4.5 improves significantly across prompt adherence, alignment, and aesthetics.
Capability
Max output resolution
Seedream 4.5
Up to 4K
Up to 4K
Capability
Prompt adherence (MagicBench)
Seedream 4.5
Significantly improved
Baseline
Capability
Multi-image composition
Seedream 4.5
Up to 10 reference images with accurate subject identification
Limited multi-image support
Capability
Reference detail preservation
Seedream 4.5
Face, lighting, and color tone strictly preserved
Strong, with occasional drift
Capability
Typography
Seedream 4.5
Dense, small, and handwritten fonts; cross-script layouts
Standard text rendering
Capability
Material transformation
Seedream 4.5
Scoped localized edits with lighting adjustment
Whole-image style changes
Capability
Aspect ratios
Seedream 4.5
1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 21:9
Same set
| Capability | Seedream 4.5 | Seedream 4.0 |
|---|---|---|
| Max output resolution | Up to 4K | Up to 4K |
| Prompt adherence (MagicBench) | Significantly improved | Baseline |
| Multi-image composition | Up to 10 reference images with accurate subject identification | Limited multi-image support |
| Reference detail preservation | Face, lighting, and color tone strictly preserved | Strong, with occasional drift |
| Typography | Dense, small, and handwritten fonts; cross-script layouts | Standard text rendering |
| Material transformation | Scoped localized edits with lighting adjustment | Whole-image style changes |
| Aspect ratios | 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3, 21:9 | Same set |
Head to the create page and choose this model from the dropdown list.
Describe the scene, style, and motion you want. Adjust settings as needed.
Click create, then download or share once the generation finishes.
Real community works and curated prompts — copy or reuse with one click.