How to run FLUX.1 Fill inpainting in ComfyUI on a cloud GPU

Rent a GPU on Deploy, then open the ComfyUI template. Follow the official ComfyUI FLUX examples page for the Fill model: put flux1-fill-dev.safetensors in ComfyUI/models/diffusion_models/, add the text encoders and VAE that page names, drag the inpaint and outpaint example images into ComfyUI, and copy the output folder off the instance before you destroy it. Destroying the instance deletes its disk.

The same examples page’s FLUX.1-dev files are 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. A different inpainting program, whose README prints no GPU GB figure, is How to run IOPaint inpainting and object removal 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 file model_licenses/LICENSE-FLUX1-dev in the upstream flux repository is titled FLUX.1 [dev] Non-Commercial License v1.1.1. Its definition of the FLUX.1 [dev] Model includes FLUX.1 Fill [dev]. Section 2 grants use of the weights only for Non-Commercial Purposes as that file defines them. Section 2(d) says the license claims no ownership of outputs, that you may use an output for any purpose, including a commercial purpose, except as the license prohibits, and that you may not use an output to train, fine-tune, or distill a model that is competitive with the FLUX.1 [dev] Model or the FLUX.1 Kontext [dev] Model. Section 2(e) requires content filters or a review for unlawful or infringing content, and a disclosure that the output was generated or modified with artificial intelligence where the law requires it. Section 4 prohibits uses that include surveillance, biometric processing, unlawful content, child sexual abuse material, and non-consensual intimate images. A commercial or production use of the weights is not granted by this text. Read the file. This page does not restate the rest of it. The LICENSE.md and the README inside the gated model repository returned HTTP 401 on 2026-11-13, so this page does not claim that file is the same text and does not quote a VRAM figure from the card.

The commands below are copied from the upstream pages named in each step. They were not run on a CloudGPU instance for this page. Anything this page adds around them is marked.

What the examples page says about memory

The ComfyUI FLUX examples page Fill section publishes no GPU GB figure (read 2026-11-13). The page says the fp16 text encoder is recommended if you have more than 32GB ram. That sentence is system RAM, not a catalog vram_gb. The same page says you can set weight_dtype to fp8 on the Load Diffusion Model node, which will lower the memory usage by half. That sentence has no GB cutoff. The Fill section does not name a separate fp8 file.

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 2026-11-13T16:03:04.945Z. GET /v1/public/gpu-catalog the same day (generated_at 2026-11-13T16:03:03.877Z) 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. There is no upstream GPU GB cutoff on the Fill section to match these values against. Which listing you can start is on Deploy. This page does not say which listings are available.

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

The catalog template whose name is ComfyUI has id flux-1-dev, recommended_gpu RTX 4090 48G, and min_vram_gb 24 (read 2026-11-13). Those three fields are the template record. They are not a Fill figure from the examples page. The model card README was not readable, as the license section says.

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 examples page does not print a shell command that finds the ComfyUI directory. This page adds the same lookup the FLUX.1-dev guide uses. It was not run for this page.

COMFY=$(find / -maxdepth 4 -type d -name ComfyUI 2>/dev/null | head -n 1); echo "$COMFY"

Files the examples page names

The extras section says the following examples might require some of the regular flux files linked at the top of that page. Those placement lines, copied:

Text encoders go in ComfyUI/models/text_encoders/. The page links t5xxl_fp16.safetensors and clip_l.safetensors, and says you can use t5xxl_fp8_e4m3fn_scaled.safetensors instead for lower memory usage, with the fp16 file recommended if you have more than 32GB ram. That link is https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main.

The VAE goes in ComfyUI/models/vae/. The page links https://huggingface.co/Comfy-Org/Lumina_Image_2.0_Repackaged/blob/main/split_files/vae/ae.safetensors.

The Fill diffusion file, copied from that section: download flux1-fill-dev.safetensors and put it in ComfyUI/models/diffusion_models/. The page links that file. This page does not copy the Hub id on the link. The model API for that repository, read 2026-11-13, reported gated: auto, cardData.license other, lastModified 2025-06-27T16:23:07.000Z, and sha 358293da0354175698b67ec8299acf928313a78a. The file listing reported flux1-fill-dev.safetensors at 23804922408 bytes and ae.safetensors at 335304388 bytes. Those are file sizes, not VRAM. The examples page tells you to take the VAE from the Comfy-Org link above. This page does not assume the two ae.safetensors files are the same bytes. Because the repository is gated, accept the license on the model page with your Hugging Face account before the download. How to download a Hugging Face model is the hf download reference. The examples page does not print an hf download command.

Workflow

The examples page says you can drag the inpaint image into ComfyUI, and that you can right click images in the Load Image node and choose Open in MaskEditor. It then shows an outpaint image to drag in the same way. This page does not restate node values that are only inside those images. The image paths on that site are /ComfyUI_examples/flux/flux_fill_inpaint_example.png and /ComfyUI_examples/flux/flux_fill_outpaint_example.png.

Refresh ComfyUI so it sees the new files. To open the page, use the public HTTPS URL the ComfyUI template shows on the instance row, or an SSH tunnel. The FLUX.1-dev guide uses a tunnel to port 8188 for this template. This page did not re-check that port. How to use SSH port forwarding is the forwarding reference.

When it fails

CUDA out of memory. The Fill section publishes no GPU GB cutoff. The examples page says you can set weight_dtype to fp8 on Load Diffusion Model, which will lower the memory usage by half. Then use how to fix CUDA out of memory on a cloud GPU.

“CUDA error: no kernel image is available” on an RTX 5090. Catalog RTX 5090 and RTX 5090D are 32. That is the catalog value, not a Fill requirement. The examples page does not pin a PyTorch build. The RTX 5090 guide says that card needs a PyTorch build for CUDA 12.8 or newer. This page did not re-test the ComfyUI template on an RTX 5090.

Gated download. The model API reported gated: auto for the repository that hosts flux1-fill-dev.safetensors. Accept the license on that model page, then download again. The license in the first section still covers FLUX.1 Fill [dev].

Save the images, then destroy

ComfyUI’s folder_paths.py sets output_directory to os.path.join(base_path, "output"). That line was read from ComfyUI HEAD 48a160ec665ff16d14469ae0f777d50eb323ca58 on 2026-11-13. The last commit that changed folder_paths.py was 12081cf9b8cb2c8128e32f689aabc8ac4688fde7, committed 2026-10-13T15:46:51Z. Copy that output directory, and any workflow you exported, 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.

Frequently asked questions

How much VRAM does the FLUX.1 Fill examples section publish?

The Fill section of the official ComfyUI FLUX examples page publishes no GPU GB figure (read 2026-11-13). The model card README returned HTTP 401 the same day, so this page does not quote a VRAM figure from it. The examples page says the fp16 text encoder is recommended if you have more than 32GB ram. That sentence is system RAM, not a catalog vram_gb. It also says setting weight_dtype to fp8 on Load Diffusion Model will lower the memory usage by half. That sentence has no GB cutoff. The catalog template named ComfyUI has min_vram_gb 24. That field is the template record, not a Fill figure. This page does not mark any catalog vram_gb as enough. Which listing you can start is on https://cloudgpu.app/deploy.

What license covers FLUX.1 Fill?

model_licenses/LICENSE-FLUX1-dev in the upstream flux repository is the FLUX.1 [dev] Non-Commercial License v1.1.1, and its definition of the FLUX.1 [dev] Model includes FLUX.1 Fill [dev] (read 2026-11-13). Section 2 limits use of the weights to Non-Commercial Purposes as that file defines them. Section 2(d) says outputs may be used for any purpose, including commercial purposes, except as the license prohibits, and that outputs may not be used to train, fine-tune, or distill a competitive model. Section 2(e) requires content filters or review, and an AI disclosure where the law requires one. The model repository LICENSE.md returned HTTP 401, so this page does not claim that file is the same text. That wording is the upstream license, not a CloudGPU price and not a storage allowance.

Where does the official ComfyUI example put the Fill weights?

The ComfyUI FLUX examples page says to download flux1-fill-dev.safetensors and put it in ComfyUI/models/diffusion_models/. It says the extras examples might require the regular flux text encoders in ComfyUI/models/text_encoders/ and the VAE in ComfyUI/models/vae/. Drag the inpaint example image and the outpaint example image on that page into ComfyUI to load the workflows. This page does not copy the Hub id on the Fill link.

Do the images stay on the instance after I destroy it?

No. The ComfyUI file folder_paths.py joins the base path with output. Copy that directory off the instance before you destroy it. The terms say data on local instance storage is permanently deleted when the instance is destroyed.

Sources (read 2026-11-13, Asia/Shanghai)

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).