Image inference with ComfyUI¶

Run image generation and image editing models in ComfyUI on RunPod. The templates provision the models, LoRAs, custom nodes and example workflows needed for:

Z-Image
FLUX.2 Dev
FLUX.2 Klein 9B
JoyCaption
Krea-2
SenseNova U1.5
Qwen-Image/Edit
Qwen-Image 2.1

What to expect¶

These templates are intended for users who want to run ComfyUI on RunPod and are comfortable following technical setup steps. No Linux expertise is required for normal use, but basic familiarity with RunPod pods, logs, tokens and file management is helpful.

The pod is an experimental workspace for image creators who want to combine generation models, editing models, LoRAs and custom nodes in open and flexible creative pipelines. It supports image generation, identity-aware editing, prompt enhancement, captioning and controlled image creation.

These pods use where possible heretic/uncensored models and custom_nodes to reduce saveguards and refusals

The approach is aligned with the open-model discussion in Eric Hartford's Uncensored Models: local and open AI systems give advanced users more ownership, control and composability. The pods are a technical workspace for responsible use rather than a content platform or safety policy.

Included features¶

Automatic model selection dependant on selected hardware.
Ready-to-use workflows for image generation, editing and enhancement.
CUDA 12.8/13.x with compiled attention and GPU acceleration.
LoRA Manager and pre-installed custom nodes.
Authentication for ComfyUI, Code Server, Hugging Face and CivitAI.
Prompt enhancers working with gpu accelerated native llama-cpp or llama-cpp-python.

RunPod deployment¶

Start here: Choose, deploy and start an image template on RunPod.

Once the pod is running, use the workflow examples below to choose what you want to create. Follow the ComfyUI tutorial for the steps from opening ComfyUI to running your first workflow.

Choose a workflow¶

Goal	Input	Recommended workflow	Go to
Preserve or transfer a subject's identity	Source and target images	Krea-2 identity workflows	Workflows
Generate an image from a description	Text prompt	Krea-2 text-to-image	Workflows
Generate or edit images with SenseNova U1.5	Text prompt, optionally with an input image	SenseNova U1.5 text-to-image and image-to-image	Workflows
Generate or edit images with Qwen-Image 2.1	Text prompt, optionally with reference images	Qwen-Image 2.1 text-to-image and image-to-image, with optional prompt enhancement	Workflows
Guide generation with a control or reference image	Prompt and control image	FLUX.2 Klein control	Workflows
Create several views of the same subject	Image and angle instructions	Multiple-angle workflows	Workflows
Generate images with ZIB-ZIT	Text prompt	ZIB-ZIT	Workflow
Describe or caption an image	Image	JoyCaption	Workflow

Krea-2¶

Example images¶

t2i ostris controlnet

t2i ostris controlnet

t2i Qwen text/image conditioning

t2i conditioning

i2i identity edit

t2i edentity edit

Identity workflows¶

Use these image-to-image workflows to transfer, edit or compose identities while retaining control over the target image.

View the Krea-2 identity transfer workflow

View the Krea-2 identity edit loop

View the Krea-2 identity composition workflow

View the Krea-2 Ostris edit workflow

text-to-image workflows¶

Choose a workflow based on how you want to build and condition the prompt.

Workflow	Best suited for
Prompt enhancer	Expanding a short prompt before generation
Image conditioning without VAE encoding	Conditioning directly from a reference image
Artist-friendly conditioning	Creative control with an artist-oriented setup

View the Krea-2 prompt enhancer workflow

Krea-2 text-to-image workflow with prompt enhancer

View the Krea-2 image-conditioning workflow

Krea-2 text-to-image workflow with image conditioning and no VAE encoding

View the Krea-2 artist-friendly workflow

Krea-2 text-to-image workflow with artist-friendly conditioning

View the Krea-2 ostris controlnet with 3D pose workflow

s

Krea-2 text-to-image workflow controlnet

SenseNova U1.5¶

Use SenseNova U1.5 for text-to-image generation or image-to-image editing with the Final BF16 checkpoint and optional distilled 8-step LoRA.

View the SenseNova U1.5 workflow

SenseNova U1.5 image workflow in ComfyUI

Qwen-Image 2.1¶

Use Qwen-Image 2.1 for text-to-image generation and image editing, with optional prompt enhancement.

Example images¶

t2i prompt enhancement phrase weight adjusted turbo lora

t2v prompt enhancement phrase weight adjusted

i2i prompt enhancement turbo lora

i2i prompt enhancement turbo lora

i2i controlnet , pose created with 3D pose custom node

i2i controlner

Workflows included¶

Workflow	Best suited for
`t2i`	Generating an image from a text prompt
`i2i`	Editing an image with text instructions
`t2i-prompt-enhancer`	Expanding a prompt before image generation
`i2i-prompt-enhancer`	Enhancing instructions before image editing
`i2i-controlnet`	Modify the pose of a reference inmage with 3D pose

View the Qwen-Image 2.1 text-to-image prompt enhancer workflow

Qwen-Image 2.1 text-to-image workflow with prompt enhancer

View the Qwen-Image 2.1 text-to-image phrase weights workflow

Qwen-Image 2.1 text-to-image workflow with phrase weights

View the Qwen-Image 2.1 image-edit prompt enhancer workflow

Qwen-Image 2.1 image-edit workflow with prompt enhancer

View the Qwen-Image 2.1 image-edit prompt enhancer custom node

Qwen-Image 2.1 prompt enhancer custom node

View the Qwen-Image 2.1 controlnet workflow with 3D poser

FLUX.2 Klein workflows¶

View the FLUX.2 Klein multiple image composition workflow

FLUX.2 Klein multiple images composition workflow

Use FLUX.2 Klein when generation needs to follow a control, target or reference image.

View the FLUX.2 Klein control and target workflow

FLUX.2 Klein control and target image generation workflow

View the FLUX.2 Klein RefControl workflow

FLUX.2 Klein RefControl image generation workflow

Multiple-angle workflows¶

Generate several viewing angles with Qwen Image 2.1 , Qwen-Image-Edit or FLUX.2 Dev.

View the Qwen-Image 2.1 mutiple angle workflow

View the Qwen-Image-Edit multiple-angle workflow

Qwen-Image-Edit workflow for multiple camera angles

View the FLUX.2 Dev multiple-angle workflow

FLUX.2 Dev workflow for multiple camera angles

ZIB-ZIT¶

Use the ZIB-ZIT workflow for text-to-image generation with the Z-Image model family.

View the ZIB-ZIT workflow

JoyCaption¶

Use JoyCaption to analyze an image and generate a detailed description or prompt.

View the JoyCaption workflow

Hardware output comparison¶

These screenshots show the workflows running with different models, precisions and GPU configurations.

Workload	Precision	Example GPUs	Note
Z-Image	—	RTX A4500, RTX A5000, RTX 4000 Ada, RTX 3090	Four GPU examples
FLUX.2 Klein 9B	—	RTX A4500	One GPU example
Qwen-Image-Edit	fp8 or bf16	RTX A5000, A40	Two precision examples
FLUX.2 Dev	bf16 or fp8	L40S, RTX A5000	The fp8 A5000 example is slower
Krea-2 turbo	fp8 or bf16	RTX A5000, RTX 4090	Three configuration examples
SenseNova U1.5	bf16	L40S	L40S example; t2i and i2i also tested on RTX 6000 Ada
Qwen-Image 2.1	INT8 ConvRot	L40S, RTX 4090,6000 PRO MiG 24 Gb	Tested with 60 GB volume storage;