Icon

Updated weekly

OW

OW

Icon

Updated weekly

OW

Output Weekly

Output Weekly

Tools, workflows and honest rankings for people who make things with AI

Tools, workflows and honest rankings for people who make things with AI

Creative Freedom

5 best uncensored image-to-image generators in 2026

Danielle Kerr

A digital painting of a man's face and torso

Key takeaways

  • Mage ranks first for a hosted tool. Mango 4 edits an input image and takes up to 10 reference images, and it runs unlimited from the $30 Pro plan. Inpaint and Refine work on any upload.

  • ComfyUI runs image-to-image and inpainting on your own graphics card, with a denoise setting that sets how much of the original survives. The model's license and the law still apply.

  • NovelAI has Strength and Noise sliders and a masked Inpaint, built around anime models, and its terms set a minimum age of 18.

  • Leonardo filters prompts and flags outputs, and Midjourney's guidelines allow safe-for-work content only. Both rank here for their editing tools, not for loose rules.

  • Mage and Midjourney both restrict edits of real people, so use images you own or original characters.

Which image-to-image generators are uncensored?

For a hosted tool, Mage is the permissive pick, with Mango 4 editing and Inpaint working on any upload. For no host filter at all, run image-to-image in ComfyUI on your own graphics card.

"Uncensored" is a loose label, and every company draws its own lines. So this ranking rests on what each tool's own rules say, then on what it can do with a photo you upload, and then on cost. Image-to-image means you hand the model a picture and it redraws part or all of it. That covers restyling, variations, reference-based edits, inpainting (repainting a masked area), and character reference. Our uncensored AI image generator roundup and text-to-image ranking cover prompt-only work.

The five tools at a glance

Tool

Where it runs

Edit tools for an upload

Starting cost

Mage

Browser

Mango 4 edits, Inpaint, Refine, up to 10 references

300 free Gems once, then $10/mo

ComfyUI

Your PC

Image-to-image, inpainting, edit models

Free with your own GPU

NovelAI

Browser

Image2Image, Inpaint, Precise Reference

30 free images, then $10/mo

Leonardo

Browser

Image to Image, Canvas inpainting, Image Guidance

150 free tokens a day, $12/mo

Midjourney

Browser and Discord

Edit model, Editor, up to 4 images

$10/mo

1. Mage: best overall for permissive image editing

Mage is a browser platform that runs its own uncensored image models next to open ones.

Strengths: Mango 4, Mage's default image model and a Mage exclusive, edits an input image, and its docs say it works on part of an image while the rest stays put. It takes up to 10 reference images, so a saved Character can carry a face into an edit. The Inpaint app repaints a masked area from a prompt on Mango, Guava, Seedream, GPT Image 2, or Nano Banana 2, and Refine runs a whole-image pass. Flux 2 takes up to 3 input images, and ControlNet works on Stable Diffusion 1.5 and XL.

Limits: Mango 4 costs Gems on the $10 Basic plan and runs unlimited at 1K and 2K from Pro at $30. Free accounts get 300 Gems once, and they expire after 90 days.

Best for: Edits and reference-based variations on a recurring character, in a browser.

2. ComfyUI: best for no host filter

ComfyUI is free software under GPL-3.0 that runs open models on your own computer, so no company reviews your uploads or prompts. Its README says it can run on as little as 4GB of video memory, and our VRAM guide covers what bigger models need.

Strengths: Its image-to-image example puts the main control in one line. "The lower the denoise the less noise will be added and the less the image will change." Inpainting works with ordinary checkpoints too. The README lists edit models including Flux Kontext, Qwen Image Edit, and Flux.2 Klein, and you can stack LoRAs and ControlNet as you like. Fixing a seed lets you change one setting at a time.

Limits: The license travels with the file. SDXL's CreativeML Open RAIL++-M license bars using the model to exploit or harm minors "in any way" and to break the law. Setup, model files, and updates are yours to manage.

Best for: People with a capable graphics card who want full control over every edit.

3. NovelAI: best for anime image-to-image

NovelAI, from Anlatan, makes anime-style images on its own models. Its Image2Image tool needs a base image, and you can drag one in, upload it, or send a generated image back as the base.

Strengths: The Strength and Noise docs say Strength "defines how much your written prompt will affect the image" and how much the AI may change the upload. Noise helps it add new detail. Inpaint repaints a masked area and has its own Strength slider on V4 and newer models. Precise Reference, on V4.5 models, guides a generation with a reference image for 5 extra Anlas.

Limits: Plans cost $10, $15, and $25 a month, and Opus's Anlas-free generations don't cover images made from a base image. The terms set a minimum age of 18 and say encrypted copies of your uploaded images may be cached temporarily, then deleted automatically.

Best for: Anime and illustration edits with fine control over how far the result drifts.

4. Leonardo: best for a mainstream editing suite

Leonardo is a browser-based image suite with a full set of tools for working from an existing image. Its Image to Image option uses a picture as the starting point at no extra token cost.

Strengths: The Canvas editor covers inpainting, outpainting, and Image to Image, and its Inpaint Input Strength setting "Adjusts how closely the new generation will follow the underlying image." Image Guidance adds Style, Content, and Character Reference, and Omni models take up to 6 images.

Limits: Leonardo's FAQ says certain word combinations can trigger prompt filters, possible NSFW output is flagged, and "Free users are subject to stricter filtering." Character Reference takes a single image and sits in a legacy mode that's leaving the platform.

Best for: Existing Leonardo users who want inpainting and reference tools in one suite.

5. Midjourney: best for polished edits under strict rules

Midjourney's Edit model, on versions 8.1 and 8.2, changes an image from written instructions and generates from up to 4 reference images. It replaces Omni Reference and Character Reference.

Strengths: The Editor handles inpainting and outpainting, and the Edit model works with Style References, Moodboards, and Personalization. Adding "--raw" reduces Midjourney's extra creative flair.

Limits: The community guidelines say content must be "Safe For Work," and an Editor-specific rule says not to use it to create NSFW or sexualized imagery. Every plan is a paid subscription, starting at $10 a month.

Best for: Stylized, high-polish edits where loose rules aren't a factor.

What each image-to-image control changes

  • Denoise or Strength: Low values keep the original nearly intact, and high values let the model reinterpret it. ComfyUI calls it denoise, NovelAI calls it Strength, and Leonardo calls it Input Strength.

  • Mask: Inpainting repaints only the area you paint over, so the rest of the image stays put.

  • Reference images: These guide a new image instead of redrawing the old one. Mage takes up to 10 on Mango 4, and Midjourney's Edit model takes up to 4.

  • ControlNet: This holds pose or edges from your upload while the model repaints everything else.

Uncensored image-to-image generator FAQ

What is the best uncensored image-to-image generator?

Mage is the first pick for most people. It's permissive by hosted standards, and Mango 4 edits an input image with up to 10 references from $30 a month. Its Terms say, "We prohibit generating, promoting, distributing, or sharing pornographic content." For no host filter, run ComfyUI locally.

What does denoise strength do?

It sets how much of your uploaded image the model keeps. A low value changes little, so it suits small touch-ups and light restyles. A high value lets the prompt dominate, which suits big changes and costs you more of the original composition.

Can I upload a photo of a real person?

Only an image you own or have the rights to, and only with the person's consent. Mage's content rules say generating a recognizable real person without consent isn't permitted, and Midjourney's Editor rules bar manipulating real people's images in ways that could lead to harassment, abuse, or harm. A local setup has no company enforcing this, but the law still applies.

Does image-to-image keep my face?

It can, if you use a reference or a saved character and keep the denoise low. Mage's Characters and References, Leonardo's Character Reference, and Midjourney's Edit model are built for this. Our LoRA explainer covers the local route, where a trained LoRA holds the face, and our character consistency ranking compares the tools.

Can I run image-to-image for free?

Locally, yes, if you already own a suitable graphics card. Online, Mage gives 300 Gems once at sign-up, NovelAI offers 30 free images, and Leonardo's free plan gives 150 tokens a day. Midjourney's plans are all paid subscriptions, starting at $10 a month.