How to run FLUX.1 Krea [dev] in ComfyUI on a cloud GPU
Rent a GPU on Deploy, then open the ComfyUI template or a terminal in JupyterLab or over SSH. Follow the official FLUX.1 Krea Dev tutorial: put the diffusion, text-encoder, and VAE files that page names into the folders it prints, accept the agreement the page names before the original-weights download, and run the workflow it says to search for as “FLUX.1 Krea Dev”. Copy the output folder off the instance before you destroy it. Destroying the instance deletes its disk.
FLUX.1 [dev] file placement on the older examples page is How to run FLUX.1-dev in ComfyUI on a cloud GPU. Instruction editing with FLUX.1 Kontext [dev] is How to run FLUX.1 Kontext in ComfyUI on a cloud GPU. Inpainting with FLUX.1 Fill [dev] is How to run FLUX.1 Fill inpainting in ComfyUI on a cloud GPU. Text-to-image whose ComfyUI knowledge page prints 24GB, about 13GB, and about 7GB is How to run Chroma text-to-image in ComfyUI on a cloud GPU. Training a FLUX LoRA is How to train a FLUX LoRA with AI Toolkit on a cloud GPU and How to train a FLUX LoRA with FluxGym on a cloud GPU. Installing ComfyUI itself is How to install ComfyUI and ComfyUI-Manager on a cloud GPU. How published FLUX and SDXL weight sizes are read: How much VRAM do I need for FLUX and SDXL. If the process runs out of memory: How to fix CUDA out of memory on a cloud GPU.
The tutorial says this model is released under the flux-1-dev-non-commercial-license and links LICENSE.md on black-forest-labs/FLUX.1-Krea-dev. The model API on 2026-11-13 reported license_name flux-1-dev-non-commercial-license, cardData.license other, and gated auto. README.md and LICENSE.md on that repository returned HTTP 401 the same day, so this page does not quote them. The public file model_licenses/LICENSE-FLUX1-dev in black-forest-labs/flux is titled FLUX.1 [dev] Non-Commercial License v1.1.1. Its definition list names FLUX.1 [dev], FLUX.1 Fill [dev], FLUX.1 Depth [dev], FLUX.1 Canny [dev], FLUX.1 Redux [dev], the Canny and Depth LoRAs, and FLUX.1 Kontext [dev]. That sentence does not name FLUX.1 Krea [dev]. This page does not treat that file as the Krea license text. Read LICENSE.md on the model page after the repository lets you. That license name is not a CloudGPU price.
The steps below are copied from the tutorial and the workflow file named in each section. They were not run on a CloudGPU instance for this page. Anything this page adds around them is marked.
What the tutorial says about memory
The tutorial says t5xxl_fp16.safetensors is recommended when your VRAM is greater than 32GB, and it labels t5xxl_fp8_e4m3fn.safetensors for low VRAM. It says flux1-krea-dev.safetensors is the original weights, and that if you have enough VRAM like 24GB you can use it. It says flux1-krea-dev_fp8_scaled.safetensors is recommended for low VRAM users. A tip on the same page says that for low VRAM users this model may not run smoothly, and that you can wait for the community to provide an FP8 or GGUF version. Those sentences are the tutorial’s. This page did not remeasure them. The model-card README returned HTTP 401, so this page does not quote a VRAM figure from the card. The older ComfyUI_examples flux README does not contain the string krea. This page does not copy that README’s fp8 sentences onto this model.
GET /v1/public/compute-capabilities does not list GPUs. It returns feature flags. The memory column is vram_gb from gpu-prices.json, updated_at 2026-11-13T22:12:44.591Z. GET /v1/public/gpu-catalog the same day (generated_at 2026-11-13T22:12:44.965Z) listed the same names and the same vram_gb values, plus Ascend 910B at 64, which is not in gpu-prices.json and is not listed here. The catalog response also includes an availability field. This page does not copy it. Every prices row has gpu_count 1. Every catalog row has min_gpu_count 1. Catalog rows whose vram_gb is 24 are the values that match the number in “like 24GB”. Catalog rows whose vram_gb is greater than 32 are the values that match “greater than 32GB”. A row at 32 is not greater than 32. This page does not mark a row as enough. Which listing you can start is on Deploy.
The catalog template whose name is ComfyUI has id flux-1-dev, recommended_gpu RTX 4090 48G, and min_vram_gb 24. Those three fields are the template record. They are not a sentence from the Krea tutorial. The template list had no template whose name contained Krea.
| Catalog name | vram_gb |
|---|---|
| RTX 2080 | 8 |
| RTX 3080 | 10 |
| RTX 2080 Ti | 11 |
| RTX 3080 Ti | 12 |
| V100 16G | 16 |
| A30 | 24 |
| RTX 3090 | 24 |
| RTX 4090 | 24 |
| RTX 4090D | 24 |
| RTX 4080 SUPER 32G | 32 |
| RTX 5090 | 32 |
| RTX 5090D | 32 |
| V100 32G | 32 |
| A100 40G | 40 |
| A800 40G | 40 |
| A40 | 48 |
| RTX 4090 48G | 48 |
| RTX 4090D 48G | 48 |
| RTX A6000 | 48 |
| NVIDIA RTX PRO 5000 72G | 72 |
| RTX PRO 5000 72G | 72 |
| A100 80G | 80 |
| A800 80G | 80 |
| H800 | 80 |
| RTX 6000D | 83 |
| H20 96G | 96 |
| RTX PRO 6000 96G | 96 |
| RTX PRO 6000 Max-Q | 96 |
| H20-3e 141G | 141 |
Install on the ComfyUI template
On Deploy, choose the template named ComfyUI. How to choose a rental GPU for ComfyUI is the card page for that template. The manual install of ComfyUI and ComfyUI-Manager is How to install ComfyUI and ComfyUI-Manager on a cloud GPU. The FLUX.1-dev guide says the template installs ComfyUI only and does not include FLUX weights, citing CloudGPU’s docs as read on 2026-11-03. This page did not re-read those docs. Read the template description on Deploy before you rely on that sentence.
The tutorial does not print a shell command that finds the ComfyUI directory. This page adds the same lookup the FLUX.1 Fill guide uses. It was not run for this page.
# Added on this page. The tutorial does not print this lookup.
COMFY=$(find / -maxdepth 4 -type d -name ComfyUI 2>/dev/null | head -n 1); echo "$COMFY"
Files the tutorial names
The tutorial’s save-location tree, copied:
ComfyUI/
├── models/
│ ├── diffusion_models/
│ │ └── flux1-krea-dev_fp8_scaled.safetensors or flux1-krea-dev.safetensors
│ ├── text_encoders/
│ │ ├── clip_l.safetensors
│ │ └── t5xxl_fp16.safetensors or t5xxl_fp8_e4m3fn.safetensors
│ ├── vae/
│ │ └── ae.safetensors
The diffusion links on that page are flux1-krea-dev_fp8_scaled.safetensors at Comfy-Org/FLUX.1-Krea-dev_ComfyUI and flux1-krea-dev.safetensors at black-forest-labs/FLUX.1-Krea-dev. The text-encoder links are clip_l.safetensors, t5xxl_fp16.safetensors, and t5xxl_fp8_e4m3fn.safetensors at comfyanonymous/flux_text_encoders. The VAE link is ae.safetensors at black-forest-labs/FLUX.1-schnell. The model API for Comfy-Org/FLUX.1-Krea-dev_ComfyUI on 2026-11-13 reported gated false and cardData.license other. lastModified was 2026-10-28T02:41:56.000Z. The model API for black-forest-labs/FLUX.1-Krea-dev the same day reported gated auto, sha 8162a9c7b05a641be098422bf2fcf335615c2f28, and lastModified 2025-07-31T14:33:39.000Z.
A note on the tutorial says the flux1-dev.safetensors file requires agreeing to the black-forest-labs/FLUX.1-Krea-dev agreement before downloading via browser. The diffusion bullet in the same section links flux1-krea-dev.safetensors, not flux1-dev.safetensors. This page does not choose between those two filenames. Because the original-weights repository is gated, accept the agreement on the model page with your Hugging Face account before that download. How to download a Hugging Face model is the hf download reference. The tutorial does not print an hf download command.
Workflow
The tutorial says to download the JSON or search “FLUX.1 Krea Dev” in the Template Library. The download link is Comfy-Org/workflow_templates templates/flux1_krea_dev.json. The latest commit that touched that file, read 2026-11-13, was ea7faae6999fb31338f43408ff4a575bd3f7d9a0, committed 2026-11-08T16:45:29Z. The file’s SaveImage node has widgets_values ["flux_krea"]. The verification list says to load flux1-krea-dev_fp8_scaled.safetensors or flux1-krea-dev.safetensors in the Load Diffusion Model node, t5xxl_fp16.safetensors or t5xxl_fp8_e4m3fn.safetensors as clip_name1 and clip_l.safetensors as clip_name2 in the DualCLIPLoader node, and ae.safetensors in the Load VAE node, then click Queue or use Ctrl(cmd) + Enter. This page does not restate node values that are only inside the preview image.
The tutorial does not document a separate Gradio launch. ComfyUI is the UI. How to install ComfyUI and ComfyUI-Manager on a cloud GPU says comfy/cli_args.py defaults --listen to 127.0.0.1 and --port to 8188. How to use SSH port forwarding on a cloud GPU is the tunnel. This page did not install the workflow and did not open the tunnel.
When it fails
CUDA out of memory. The tutorial’s memory lines are the 32GB text-encoder sentence, the “like 24GB” original-weights sentence, and the low-VRAM fp8 labels. It does not print a further cutoff. If the process raises that error, use how to fix CUDA out of memory on a cloud GPU.
“CUDA error: no kernel image is available for execution on the device” on an RTX 5090. The tutorial does not pin a PyTorch build and does not name compute capability 12.0. The RTX 5090 guide says that card is compute capability 12.0 (sm_120) and that PyTorch 2.7 and later, on a build for CUDA 12.8 or newer, is the line that announced support. This page did not install a torch wheel and did not re-test this workflow on an RTX 5090. Catalog RTX 5090 and RTX 5090D are 32. That is the catalog number only. It is not greater than 32.
Save the images, then destroy
The workflow’s SaveImage prefix is flux_krea. The install guide says folder_paths.py sets the output directory to the folder named output. Copy that directory off before you destroy the instance. Save your work before you destroy a cloud GPU is the checklist. The terms say data on local instance storage is permanently deleted when the instance is destroyed.
The pricing page FAQ states the billing rule in these words:
Per-minute options have a 1-minute minimum; unused prepaid minutes are refunded when you destroy the instance. Some configurations use whole-hour billing with a 1-hour minimum; unused time within a charged hour is not refunded on voluntary stop. The billing period, minimum charge and initial prepayment are shown before confirmation. No subscription.
Prices are not copied here.
Frequently asked questions
How much VRAM does the FLUX.1 Krea tutorial print?
The ComfyUI tutorial says t5xxl_fp16.safetensors is recommended when your VRAM is greater than 32GB, and that flux1-krea-dev.safetensors is the original weights if you have enough VRAM like 24GB. It recommends flux1-krea-dev_fp8_scaled.safetensors for low VRAM users and labels t5xxl_fp8_e4m3fn.safetensors for low VRAM. A tip on the same page says low VRAM users may not run smoothly and can wait for a community FP8 or GGUF version. The model card README returned HTTP 401, so this page does not quote a VRAM figure from it. This page does not mark a catalog row as enough. Which listing you can start is on https://cloudgpu.app/deploy.
Which files does the tutorial put in which folders?
The tutorial’s tree puts flux1-krea-dev_fp8_scaled.safetensors or flux1-krea-dev.safetensors in ComfyUI/models/diffusion_models/, clip_l.safetensors and either t5xxl_fp16.safetensors or t5xxl_fp8_e4m3fn.safetensors in ComfyUI/models/text_encoders/, and ae.safetensors in ComfyUI/models/vae/. The fp8 diffusion link is Comfy-Org/FLUX.1-Krea-dev_ComfyUI. The original weights link is black-forest-labs/FLUX.1-Krea-dev. The VAE link is ae.safetensors on black-forest-labs/FLUX.1-schnell. A note on the page says flux1-dev.safetensors requires agreeing to the black-forest-labs/FLUX.1-Krea-dev agreement. This page did not download the files.
Where are the images saved?
The workflow template templates/flux1_krea_dev.json has a SaveImage node whose widgets_values are flux_krea. The ComfyUI install guide says folder_paths.py sets the output directory to the folder named output. This page did not run the workflow.
What license does the tutorial name?
The tutorial says this model is released under the flux-1-dev-non-commercial-license and links LICENSE.md on black-forest-labs/FLUX.1-Krea-dev. The model API on 2026-11-13 reported license_name flux-1-dev-non-commercial-license, cardData.license other, and gated auto. README.md and LICENSE.md returned HTTP 401. The public LICENSE-FLUX1-dev file’s definition list does not name FLUX.1 Krea [dev]. That license name is not a CloudGPU price.
Sources (read 2026-11-13, Asia/Shanghai)
- ComfyUI tutorial
https://docs.comfy.org/tutorials/flux/flux1-krea-dev(folders, fp8 and original filenames, greater than 32GB, like 24GB, low-VRAM labels, the community-FP8 tip, the agreement note, the flux-1-dev-non-commercial-license sentence, search text “FLUX.1 Krea Dev”): markdown fetched 2026-11-13 Comfy-Org/workflow_templatestemplates/flux1_krea_dev.json(SaveImagewidgets_valuesflux_krea): latest commit touching the fileea7faae6999fb31338f43408ff4a575bd3f7d9a0, committed 2026-11-08T16:45:29Z- Hugging Face model API
black-forest-labs/FLUX.1-Krea-dev(gatedauto,license_nameflux-1-dev-non-commercial-license,cardData.licenseother,lastModified2025-07-31T14:33:39.000Z,sha8162a9c7b05a641be098422bf2fcf335615c2f28).README.mdandLICENSE.mdreturned HTTP 401 - Hugging Face model API
Comfy-Org/FLUX.1-Krea-dev_ComfyUI(gatedfalse,cardData.licenseother,lastModified2026-10-28T02:41:56.000Z) black-forest-labs/fluxmodel_licenses/LICENSE-FLUX1-dev(title FLUX.1 [dev] Non-Commercial License v1.1.1; definition list does not name FLUX.1 Krea [dev]): read 2026-11-13comfyanonymous/ComfyUI_examplesflux/README.md(nokreastring): read 2026-11-13- Public GPU memory: https://cloudgpu.app/gpu-prices.json (
updated_at2026-11-13T22:12:44.591Z) and https://cloudgpu.app/v1/public/gpu-catalog (generated_at2026-11-13T22:12:44.965Z).GET /v1/public/compute-capabilitiesreturned feature flags and no GPU list. The catalog had no Krea template
Cards you can start are on the Deploy page. Prices are on https://cloudgpu.app/pricing and are not repeated here.
Last updated: 2026-11-13 (Asia/Shanghai).