Microsoft: TRELLIS.2
microsoft/trellis-2Hugging Face3Denimage-to-3dtrellis2
Budget for the whole asset
At a $0.04 textured-generation rate, budget roughly $0.07–$0.14 for an asset with repair. This planning example uses nine completed repair trials with DeepSeek and GLM: repair model calls and machine time added about $0.03–$0.09. It is a small test sample, not a quote or a spending limit. Other models, settings and repeated attempts can cost more.
Texture is included in the selected generation tariff; choosing textured geometry can select a higher rate. Optional repair adds the vision model’s usage and a machine charged per started minute. Usage is chargeable even if repair fails or keeps the original unchanged. The model spend ceiling excludes generation and machine time.
Allow tens of minutes: plan for 20–60+ minutes with repair. Generation and repair run in sequence; complex assets and queues can take longer. The repair time limit starts after its queue and includes machine startup, not generation. Machine cleanup can add billable time.
How 3D billing worksReference inputs & limits
Accepted uploads and guidance from each provider. Limits apply to one generation.
ImpCC · mesh generation · up to 1 file · 20 MiB total
image as reference (up to 1) · image/png, image/jpeg, image/webp
One view of the object on a plain background. A prompt with no view is rendered by the text-to-image stage first. A prompt beside a view is refused because the mesh model cannot read both.
Providers
Live endpoints for this model. Response is the time from request to result; uptime is measured over the last day and week. Nothing here streams, so there is no first token to time and no token rate to report.
| Provider | Faces | Per mesh | Response | Capacity | Uptime 1d | Uptime 7d | Health | Data |
|---|---|---|---|---|---|---|---|---|
| ImpCCint890% offCA | 2,000 – 500,000 faces · textured | $0.0040 | 632.6s1d | 1/1 free | 100.0% | 98.9% | Healthy | ZDR |
Pricing
USD per delivered mesh.
| Provider | Per mesh |
|---|---|
| ImpCCint890% off | $0.0040 |
Price history
Price · USD per delivered mesh
— No change to this price has been recorded yet.
Performance
Percentiles cover the labelled window: the last 30 minutes when it has enough samples, otherwise the last day. Response is the time in seconds from request to result. This model answers whole rather than streaming, so there is no first token to time and no rate between tokens to report — response time is the whole measurement there is.
| Provider | Response p50 | p75 | p90 | p99 |
|---|---|---|---|---|
| ImpCCint8 | 632.6s1d | 832.6s | 1113.8s | 1575.6s |
Uptime
Share of successful requests and probes over the last day and week.
- ImpCCint81d 100.0%7d 98.9%Healthy
Availability by hour · last 7 days
Share of successful requests and probes per hour and provider. The line spans hours with nothing to measure; where samples are sparse, a dot marks each hour actually measured.
Activity
Mesh jobs per UTC day over the last 30 days. Every billed request counts, the platform's own paid probe traffic included.
job=$(curl -s https://api-staging.impossiblecarrot.cc/api/v1/meshes \ -H "Authorization: Bearer $HEIMDALL_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "microsoft/trellis-2", "prompt": "a worn leather armchair, a studio prop", "target_faces": 20000, "generate_texture": true }')job_id=$(echo "$job" | jq -r '.id')# Poll until the job reaches a terminal status.while true; do status=$(curl -s https://api-staging.impossiblecarrot.cc/api/v1/meshes/$job_id -H "Authorization: Bearer $HEIMDALL_API_KEY") state=$(echo "$status" | jq -r '.status') [ "$state" = "pending" ] || [ "$state" = "in_progress" ] || break sleep 5doneecho "$status" | jq -r '.meshes[0].faces'curl -s "https://api-staging.impossiblecarrot.cc/api/v1/meshes/$job_id/content?index=0" \ -H "Authorization: Bearer $HEIMDALL_API_KEY" \ -o out.glbSet HEIMDALL_API_KEY to a key from your account. See the quickstart for routing suffixes and provider preferences.
About
One photograph of an object becomes a textured GLB mesh. The encoder reads a single view and places no second camera, so one view is what this entry takes; a prompt with no view is rendered into a view first and the mesh is built from that. `target_faces` sets the face budget, from 2,000 to 500,000, and a request naming none is decimated to 20,000. The budget is a maximum, not a promised face count or quality score. Lower budgets can lose silhouette detail; textures remain available. Use `generate_texture: false` when only geometry is needed.
- Author
- Microsoft
- Family
- microsoft/trellis-2
- Weights
- microsoft/TRELLIS.2-4B
- Parameters
- generate_texture (true/false), guidance_scale (0–20), input_references (list), negative_prompt (any value), seed (0–2147483647), steps (8–50), target_faces (2000–500000), texture_size (1024, 2048, 4096), voxel_resolution (1024, 1536)
- Licence
- mit
- Size
- —
- Tokenizer
- —
- Available until
- —
From the model card
What the repository states about these weights, read from microsoft/TRELLIS.2-4B on Oct 11, 2026.
- Languages
- English
- Papers
- 2512.14692
- Datasets
- —
- Revision
- af44b45Dec 27, 2025
- Downloads
- 1.9M
- Likes
- 1.3K