Seedream 5.5 Layerize vs Seedream 5.0 Pro Layerize: occlusion order and 20+ Live2D parts
Our image-to-layers app splits flat artwork and product visuals into transparent PNGs and PSD files. On November 13, 2026 we tested ByteDance’s Seedream 5.5 Layerize against Seedream 5.0 Pro Layerize on our own AI-generated product ad and original VTuber character before upgrading production.
Our take: Seedream 5.5 Layerize is both better and cheaper: it fixed the front-to-back stacking bug on our soda-can ad, raised max_images from 17 to 100 so a 22-part Live2D rig comes back in one call (23 layers total, with a clean empty background), and lowered per-layer pricing on KeelRouter from $0.023 to $0.02 at 1K ($0.046 to $0.036 at 2K). Complex anime rigs can still need a quick layer-stack drag in Photoshop or Live2D—on our 23-layer run, front bangs landed below the face layer—so treat z_index on 20+ part characters as a starting order rather than a final rig. We switched production to Seedream 5.5 Layerize.
Models: seedream-5.5-layerize, seedream-5.0-pro-layerize. Tested 2026-11-13, auto_1K transparent PNG layers; billed per returned layer (including base), same prompts and inputs on both.
Results
| Model | Succeeded | Average time | Average cost per run |
|---|---|---|---|
| Seedream 5.5 Layerize | 1 / 1 | 113 s | $0.22 |
| Seedream 5.0 Pro Layerize | 1 / 1 | 128 s | $0.25 |
Price on KeelRouter
| Option | Seedream 5.5 Layerize | Seedream 5.0 Pro Layerize |
|---|---|---|
| 1k per layer | $0.02 | $0.023 |
| 2k per layer | $0.036 | $0.046 |
per layer. Prices follow our catalog and update when providers change theirs.
What changed between 5.0 Pro Layerize and 5.5 Layerize
The biggest practical upgrade is raising max_images from 17 (base plus up to 16 layers) to 100 (base plus up to 99 layers). For posters, product ads and flat lays, 5 to 15 layers is already plenty, so keeping max_images at 17 by default avoids paying for tiny redundant fragments. When a prompt explicitly lists 20 or more rigging parts—left and right eyebrows, left and right eyes, upper lip, mouth interior, hair locks and clothing trims—raising max_images to match the part list lets Seedream 5.5 separate every piece in one pass.
Background infill also improved on character splits: on our VTuber test, Seedream 5.0 Pro left a ghost copy of the hoodie painted onto layer 0, whereas Seedream 5.5 left layer 0 completely blank when asked and filled in the forehead behind the bangs and the hoodie behind both sleeves.
| Capability / Tier | Seedream 5.5 Layerize | Seedream 5.0 Pro Layerize |
|---|---|---|
| 1K price on KeelRouter | $0.02 / layer | $0.023 / layer |
| 2K price on KeelRouter | $0.036 / layer | $0.046 / layer |
| max_images range (includes base) | 2 – 100 (default 17) | 2 – 17 (default 17) |
| Soda-ad occlusion order (11 layers) | Front lime stays in front of can | Front lime placed behind can |
| VTuber character split | 23 layers (clean base + 22 parts) | 16 layers (merged eyes/hair, ghost torso on base) |
Soda can ad with splashes, limes and ice: automatic split with an empty prompt
AI-generated product ad from our built-in test set (1024 × 1024). Both models ran with an empty prompt at auto_1K and default max_images (17), and each output below is recomposed strictly in the z_index order the model returned.
(empty prompt — automatic layer decomposition, max_images: 17)


113 s · $0.22 · 11 layers; can at z_index 2, front-left lime wedge above it so occlusion matches the input

128 s · $0.25 · 11 layers; front-left lime wedge returned below the can in z_index, hiding half the slice
Original VTuber character split into 23 Live2D layers in one call (max_images = 30), Seedream 5.5 only
Original anime character generated in our ChatGPT account. Seedream 5.5 Layerize raises max_images from 17 to 100; asking for 22 named Live2D parts with max_images: 30 returned a clean empty background (layer 0) plus all 22 parts on separate transparent PNGs. One z_index quirk remained: Front bangs (layer 2) came back below Face with ears (layer 8), requiring one drag above the face and eyebrows in an editor.
Split this character into these 22 Live2D layers, one part each: back hair; front bangs; left side hair; right side hair; braided side lock; star hair clip; hair ribbon; face with ears; left eyebrow; right eyebrow; left eye; right eye; upper lip and mouth line; open mouth interior with tongue; neck; hoodie body with pocket; sailor collar; chest bow ribbon; hoodie drawstrings; left sleeve and hand; right sleeve and hand; pleated skirt. Fill in hidden areas behind each part. Leave the base image as plain empty background.


216 s · $0.46 · 23 layers (empty base + 22 named parts): left/right eyebrows, left/right eyes, lip line and tongue all separated
The same VTuber character under the 17-layer ceiling, Seedream 5.0 Pro only
Shown for comparison from our October 14 rigging test on the same character. Because Seedream 5.0 Pro Layerize caps max_images at 17 (base plus at most 16 foreground layers), both eyes merged onto layer 13, bangs and side locks merged onto layer 11, the star clip and ribbon merged onto layer 15, and layer 0 kept a duplicate copy of the hoodie.
Separate the character into parts for 2D rigging (Live2D or Spine), each part on its own layer: back hair, front hair, face, eyes, mouth, neck, torso, left arm, right arm, left leg, right leg, and each accessory. Fill in what is hidden behind each part.


120 s · $0.37 · 16 layers; hit the 17-image cap, merged left/right eyes and left a duplicate hoodie on layer 0
When to use which
Seedream 5.5 Layerize
- Lower price at both resolutions: $0.02 at 1K (-13%) and $0.036 at 2K (-22%) per layer on KeelRouter.
- Supports max_images up to 100 (up to 99 foreground layers plus base) for Live2D, Spine and dense UI layouts.
- Better z_index occlusion ordering on overlapping props (kept the foreground lime wedge in front of the soda can).
- Cleaner background infill behind characters without leaving duplicate clothing on layer 0.
- 20+ layer runs take longer (216 s for 23 layers vs 113 s for 11 layers) and scale in cost with the number of layers returned.
- On 20+ part anime splits, individual hair layers (such as front bangs) can still return at the wrong z_index and need manual reordering.
Seedream 5.0 Pro Layerize
- Kept as a backward-compatible model ID for existing integrations pinned to 5.0 Pro Layerize.
- Separates 4 to 12 object product ads, flyers and sticker sheets reliably at default settings.
- Hard ceiling of max_images = 17 forces left/right eyes, eyebrows and hair locks to merge on rigging prompts.
- More likely to invert z_index on foreground props that overlap the main subject.
- Costs 15% to 28% more per layer than Seedream 5.5 Layerize.
FAQ
- Is Seedream 5.5 Layerize cheaper than Seedream 5.0 Pro Layerize?
- Yes. On KeelRouter, seedream-5.5-layerize is $0.02 per returned layer at 1K and $0.036 at 2K, down from $0.023 and $0.046 on seedream-5.0-pro-layerize.
- How many layers can Seedream 5.5 Layerize return?
- Up to 100 images per call (the base background image plus up to 99 transparent foreground layers) by passing max_images between 2 and 100. If you omit max_images, it defaults to 17.
- How is layerize billed on KeelRouter?
- Billing follows the actual number of images returned by the upstream provider (the base layer plus each foreground layer), not the max_images ceiling you pass in. An 11-layer 1K split on Seedream 5.5 Layerize bills 11 layers ($0.22), even when max_images is 17.
- Are the layers ready for Live2D without manual editing?
- Seedream 5.5 Layerize gets much closer than 5.0 Pro because it can put left/right eyes, eyebrows, lip line, mouth interior and individual hair locks on separate PNGs with hidden skin filled in. You should still expect to check layer stacking order in Photoshop or Live2D—for example, dragging front bangs above the face layer.
More comparisons
- Wan 3.0 vs Seedance 2.5: one photo, one prompt, side by side
- Kling 4.0 vs Seedance 2.0 Mini: a pug takes the clapping cat’s place
- Kling 4.0 vs Higgsfield Genjutsu: does the swap respect height?
- Claude Fable 5.2 vs Opus 5.5: price per token vs cost per task
- Gemini 4 Carbon vs Claude Opus 5.5: benchmarks, cost per task and routing
- Seedream 5.5 vs Seedream 5.0 Flash: editing curved, multi-line and RTL text in images
KeelRouter is an independent service and is not affiliated with the makers of these models. Model names are trademarks of their owners, used here only to name the model a request runs on. Results are from our own runs on the date above and may differ for other prompts.