Revision: 71153311d3dbb46851df1931d3ca6e939de83304
Stability AI model license; review current terms and commercial eligibility. Review the pinned model card and access terms before downloading.
Reviewed guide · no hardware test recorded
Generate images/video
Start with a 512-pixel SDXL-Turbo image workflow in ComfyUI. Video recipes are not yet reviewed.
Version 2026-09-27.1
comfyui-sdxl-turbo · reviewed 2026-09-27 · review due 2026-10-27
Requirements
- Total download
- Unknown
- Host disk
- Unknown
- Host RAM
- Unknown
- Per-device VRAM
- Unknown
Task details and compatibility
Exact environment sizes and per-device peak memory are not measured. Provider template disk defaults are not a verified requirement.
- task
- text-to-image
- width
- 512
- height
- 512
- batchSize
- 1
- steps
- 1
- cfg
- 1
Changing task dimensions invalidates a tested-fit claim for a different task. This version has 0 recorded successful runs.
- ComfyUI: 0.3.60. Core nodes only for the supplied API workflow. ComfyUI Manager is optional; no custom node packages are required.
- PyTorch and CUDA: version not validated. Provider image changes independently; verify the installed dependency versions and GPU architecture.
Artifacts and prerequisites
Revision: b8730510db30c8858e1e5d8e126ef19eac395560
GPL-3.0. Public source at v0.3.60. Provider template may install a different version; check before use.
Revision: 2026-09-27.1
MIT. Download and review this workflow. Uses core nodes only; it is not a hardware-tested workflow.
Set up the task
On-demand Pod with the standard ComfyUI template. Blackwell requires a separate template and is outside this recipe.
- Open the standard ComfyUI Pod template and select one supported non-Blackwell GPU. Check the environment against the pinned source version; the template itself is mutable.
- Download the pinned SDXL-Turbo FP16 checkpoint into the model checkpoint directory. Keep the exact filename used by the supplied workflow.
- Review the supplied core-node API workflow, or follow the provider's SDXL-Turbo UI template instructions. Use 512 × 512, one image, one sampling step and CFG 1 for this task.
- Queue a trial image and record runtime versions and any failure. Increasing resolution, batch size or using a video model defines a different task.
Cold download, environment setup and model loading are unknown until measured; enter a total startup allowance in the planner.
Save outputs and shut down
- Store the checkpoint in ComfyUI/models/checkpoints on a confirmed persistent mount.
- Ensure ComfyUI/output is on persistent storage, or export before stopping.
- Save generated images and the exact workflow JSON.
- Export outputs and verify local files before stopping or deleting the Pod.
- Review the selected disk's persistence and continuing charges in the lifecycle panel.
NVGPU does not provision, stop or delete resources.
Plan before you launch
Limits and compatibility
- Reviewed guide and workflow source; no GPU run recorded.
- Mutable template must be checked against the recipe before launch.
- Video workloads and Blackwell configurations are not covered by this version.
The provider can update this template independently of the pinned recipe. Check its installed versions before launching.
Sources and review evidence
Official source: Standard versus Blackwell templates, SDXL-Turbo setup and exact provider template destination.
Official source: Pinned checkpoint identity and model card.