Image Edit Feature (Fixes #1352) by Phqen1x · Pull Request #1439 · lemonade-sdk/lemonade

Phqen1x · 2026-03-25T02:04:22Z

I have added functionality to the electron app/webapp to use the /edits endpoint built into lemonade-server. Now the image generator panel has the ability to edit reference images as well using a prompt.

Copilot

Pull request overview

Adds image edit (and variation-from-reference) support to the Electron/web app by wiring the UI to the server’s /images/edits (and /images/variations) endpoints, and updates the sd-server backend behavior for variations to use the edits endpoint with embedded inference parameters.

Changes:

Update SDServer::image_variations to forward to sd-server’s /v1/images/edits and embed steps/cfg_scale via <sd_cpp_extra_args>.
Extend the Image Generation panel UI with mode selection, reference-image upload/preview, and new “Edit” / “Variations” actions on generated images.
Add/adjust CSS for new image action buttons and reference upload UI; update package-lock.json.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 5 comments.

File	Description
src/cpp/server/backends/sd_server.cpp	Changes variations forwarding behavior and prompt construction for sd-server multipart requests.
src/app/src/renderer/components/panels/ImageGenerationPanel.tsx	Adds image edit/variation workflows, reference image handling, and updated send/save behavior.
src/app/styles.css	Introduces styles for new image action buttons and reference upload UI.
src/app/package-lock.json	Lockfile update (adds transitive dependency metadata).

Files not reviewed (1)

src/app/package-lock.json: Language not supported

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot

Pull request overview

Copilot reviewed 3 out of 4 changed files in this pull request and generated 2 comments.

Files not reviewed (1)

src/app/package-lock.json: Language not supported

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

ramkrishna2910 · 2026-03-25T22:00:44Z

@Phqen1x I tried using this in the webapp, I am using SD-Turbo and when I provide a prompt with an image to edit I am seeing a new image being generated rather than the passed image being edited. See below:

Phqen1x · 2026-03-25T23:40:17Z

@Phqen1x I tried using this in the webapp, I am using SD-Turbo and when I provide a prompt with an image to edit I am seeing a new image being generated rather than the passed image being edited. See below:

I've run into this as well, but at the time I just assumed it was because the model I was accidentally using, SD-1.5, was simply too small and not powerful enough to do a good job. I just asked Claude what it thought was the issue, and it believes that the reason is because sd-server maybe uses different endpoints for reference images that isn't being used here. I am not really sure how to find what endpoint in sd-server needs to be used for the SD-Turbo, SDXL-Turbo, and other SD models to work. I know the Flux-2-Klein models do work, though.

sawansri · 2026-03-26T05:18:55Z

@Phqen1x I tried using this in the webapp, I am using SD-Turbo and when I provide a prompt with an image to edit I am seeing a new image being generated rather than the passed image being edited. See below:

I've run into this as well, but at the time I just assumed it was because the model I was accidentally using, SD-1.5, was simply too small and not powerful enough to do a good job. I just asked Claude what it thought was the issue, and it believes that the reason is because sd-server maybe uses different endpoints for reference images that isn't being used here. I am not really sure how to find what endpoint in sd-server needs to be used for the SD-Turbo, SDXL-Turbo, and other SD models to work. I know the Flux-2-Klein models do work, though.

If the issue lies in sd-server endpoints that are not implemented/unsupported, you can probably place a proxy in between the image editor and backend to capture the exact payload the editor is sending to sd-server in order to debug. Additionally, look for errors in the backend with debug logs enabled (--log-level debug).

I had to do this to reverse engineer/implement the requests claude code was sending via Anthropic's Messages API to the llamacpp backend.

I can share the script I used for the proxy if you are interested.

Phqen1x · 2026-03-26T14:16:48Z

@Phqen1x I tried using this in the webapp, I am using SD-Turbo and when I provide a prompt with an image to edit I am seeing a new image being generated rather than the passed image being edited. See below:

I've run into this as well, but at the time I just assumed it was because the model I was accidentally using, SD-1.5, was simply too small and not powerful enough to do a good job. I just asked Claude what it thought was the issue, and it believes that the reason is because sd-server maybe uses different endpoints for reference images that isn't being used here. I am not really sure how to find what endpoint in sd-server needs to be used for the SD-Turbo, SDXL-Turbo, and other SD models to work. I know the Flux-2-Klein models do work, though.

If the issue lies in sd-server endpoints that are not implemented/unsupported, you can probably place a proxy in between the image editor and backend to capture the exact payload the editor is sending to sd-server in order to debug. Additionally, look for errors in the backend with debug logs enabled (--log-level debug).

I had to do this to reverse engineer/implement the requests claude code was sending via Anthropic's Messages API to the llamacpp backend.

I can share the script I used for the proxy if you are interested.

I would appreciate if you could send that script! Thanks for the input, Sawan.

sawansri · 2026-03-26T15:19:15Z

@Phqen1x I tried using this in the webapp, I am using SD-Turbo and when I provide a prompt with an image to edit I am seeing a new image being generated rather than the passed image being edited. See below:

I've run into this as well, but at the time I just assumed it was because the model I was accidentally using, SD-1.5, was simply too small and not powerful enough to do a good job. I just asked Claude what it thought was the issue, and it believes that the reason is because sd-server maybe uses different endpoints for reference images that isn't being used here. I am not really sure how to find what endpoint in sd-server needs to be used for the SD-Turbo, SDXL-Turbo, and other SD models to work. I know the Flux-2-Klein models do work, though.

If the issue lies in sd-server endpoints that are not implemented/unsupported, you can probably place a proxy in between the image editor and backend to capture the exact payload the editor is sending to sd-server in order to debug. Additionally, look for errors in the backend with debug logs enabled (--log-level debug).
I had to do this to reverse engineer/implement the requests claude code was sending via Anthropic's Messages API to the llamacpp backend.
I can share the script I used for the proxy if you are interested.

I would appreciate if you could send that script! Thanks for the input, Sawan.

Here you go!
proxy.py

ramkrishna2910 · 2026-03-26T19:19:57Z

We could add an edit label to server_models.json and disable the edit UI for non-editing models. So users dont fail by trying SD models.

kenvandine · 2026-03-26T21:38:41Z

We could add an edit label to server_models.json and disable the edit UI for non-editing models. So users dont fail by trying SD models.

I like that solution, surely this isn't the only case that will bite us.

ramkrishna2910 · 2026-03-27T18:49:19Z

@Phqen1x let me know if you want to discuss this further. Really looking forward to the feature :)

Phqen1x · 2026-03-27T20:37:30Z

@Phqen1x let me know if you want to discuss this further. Really looking forward to the feature :)

Definitely! The edit label sounds like a great idea. I can work on implementing an edit label and integrating that into this feature tomorrow! I'll let you know if I have any other questions or comments.

Phqen1x · 2026-03-28T20:50:52Z

@ramkrishna2910, just implemented the edit label and applied it to the Flux-2-Klein models. None of the other image models included in Lemonade seem to be capable of image editing at this time. It might be nice to display the edit label in the model manager, but we're not displaying anything for the SD models now anyway. Maybe we can do that as a followup PR.

ramkrishna2910 · 2026-03-28T21:09:06Z

Yup makes sense. I will take a look at this shortly. :)

ramkrishna2910 · 2026-03-30T19:49:01Z

That worked very well!

ramkrishna2910 · 2026-03-30T19:53:46Z

The coherence is so nice!
Input image:

Prompt - Change the graph to blue

Output image:

ramkrishna2910

Great contribution! Super excited for this feature to ship!

Phqen1x · 2026-03-30T20:17:09Z

Glad to hear it! Can't wait to see it shipped as well!

kenvandine requested a review from Copilot March 25, 2026 02:30

Copilot started reviewing on behalf of kenvandine March 25, 2026 02:31 View session

Copilot AI reviewed Mar 25, 2026

View reviewed changes

kenvandine requested a review from Copilot March 25, 2026 15:13

Copilot started reviewing on behalf of kenvandine March 25, 2026 15:14 View session

Copilot AI reviewed Mar 25, 2026

View reviewed changes

Comment thread src/cpp/server/backends/sd_server.cpp

Comment thread src/app/src/renderer/components/panels/ImageGenerationPanel.tsx

jeremyfowers requested a review from ramkrishna2910 March 25, 2026 15:36

jeremyfowers assigned Phqen1x Mar 25, 2026

Phqen1x added 5 commits March 28, 2026 16:32

feat: Added integrated image editing

00f0356

fix: Commented out variations mode & improved dropdown

1890c9e

fix: Fixed suggested changes from Copilot's review

b91629e

fix: Updated comment on variations label

842a33e

fix: Edit mode only available for models that are capable of it

d845b97

Phqen1x force-pushed the image-edit branch from 9bd8ef2 to d845b97 Compare March 28, 2026 20:45

ramkrishna2910 approved these changes Mar 30, 2026

View reviewed changes

ramkrishna2910 enabled auto-merge March 30, 2026 20:03

ramkrishna2910 added this pull request to the merge queue Mar 30, 2026

Merged via the queue into lemonade-sdk:main with commit 4fa6064 Mar 30, 2026
52 checks passed

Conversation

Phqen1x commented Mar 25, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

ramkrishna2910 commented Mar 25, 2026

Uh oh!

Phqen1x commented Mar 25, 2026

Uh oh!

sawansri commented Mar 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Phqen1x commented Mar 26, 2026

Uh oh!

sawansri commented Mar 26, 2026

Uh oh!

ramkrishna2910 commented Mar 26, 2026

Uh oh!

kenvandine commented Mar 26, 2026

Uh oh!

ramkrishna2910 commented Mar 27, 2026

Uh oh!

Phqen1x commented Mar 27, 2026

Uh oh!

Phqen1x commented Mar 28, 2026

Uh oh!

ramkrishna2910 commented Mar 28, 2026

Uh oh!

ramkrishna2910 commented Mar 30, 2026

Uh oh!

ramkrishna2910 commented Mar 30, 2026

Uh oh!

ramkrishna2910 left a comment

Choose a reason for hiding this comment

Uh oh!

Phqen1x commented Mar 30, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

sawansri commented Mar 26, 2026 •

edited

Loading