2025-07-19: reForge development will continue (for now). #377
Replies: 25 comments 35 replies
|
For now I have done the backend changes to upstream from Comfy. Please report if you find new issues, and is highly factible with 150+ commits. |
|
Thank you for your work on reForge! It is my favorite UI for t2i work and I'm so glad to see you continuing to develop it. |
|
Seriously, thank you for your work. I feel not comfortable with ComfyUI (hah...) and your continued development of the WebUI client is such a blessing. |
|
Impresionante!! Gracias por tu enorme trabajo!! |
|
Welcome back! Still actively using reForge - it's still the best for SDXL! It allows wild combinations to produce original work - e.g. a combination of CFG++ sampler, BETA scheduler and RAUNet, or a combination of DPM++ 3M DY BETA with RAUNet and Smoothed Energy Guidance - none of these can be easily used in any other tool. |
|
Hi, I don't suppose I can get a more detailed explanation on how to update reForge? I see the command to use on the main page but I have no idea where to execute it, where to execute it, if I need to type anything else, etc. I wanna be able to update reForge without having to redownload the whole thing and I'm not experienced enough with Git or Python or even the CMD to know exactly what to do when I see it. Appreciate any help! |
|
Try using Stability Matrix to manage and update SDXL instances easier. You will need to scroll all the way down to the bottom to find reForge in the package lists but it holds your hand very nicely and handles all the heavy work of installing packages and updating packages and other models for you. |
|
Awesome! great to have you back |
|
Glad you are back. reForge was always one of my favourite. |
|
Hello. Still no new models. I'm relearning some things and such and trying at small steps since patching it to A1111 system is really hard. The code it is lm the backend because comfy has it on their backend. |
|
Yeah i understand it must be hard. Thanks for your work on reForge so far. Hope newer models like cosmos gets support sooner or later. |
|
Great to have you back. |
|
Glad to have you back! I tried to move to Comfy UI but it doesn't have an easy way to do batch count (not batch size) when dealing with scheduled tasks. It's super annoying that it doesn't even exist if you wanted to have a few separate image generation setups and you want each one to be run in order and have different image counts (again, not batch size in the empty latent image node, that generates multiple at the same time). |
|
I'm so glad you're back! First, I apologize for my poor English skills. English is not my first language. I wanted to let you know that your masterpiece has made my work much easier. I’m writing a web novel and needed illustrations, but my ability to create them myself was very limited, so reForge was a huge help in creating the initial drafts. :) My graphics card isn’t very powerful, so I didn’t have any other alternatives. Due to the lack of graphic resources, Forge also didn’t run properly. Thank you again. I found a few bugs while using it. In particular, I tried everything to fix the openpose issue in ControlNet, but nothing worked. I will post a detailed issue report about the relevant bugs soon! I'm really glad you're resuming this work. I'll keep cheering you on. |
|
Welcome back! |
|
First of all, welcome back @Panchovix and thanks for your hard work thus far (and in the future). reForge has been my go to for a while but I'm worried it's grown a bit too bloated (which could increase the difficulty in updating and the risk of breaking it) and in that sense I like what Classic is trying to do so I'm wondering if it wouldnt be better to build from Classic rather than Forge Gradio4 (Gradio4 is overrated). |
|
Hey that's nice! Didn't knew about neo. I have been suffering so much trying to add flux and wan support to reforge that my brain is fried. Have you managed to make chroma and Kontext work with the ldm_patched backend? I found that when adding some new architectures everything breaks a little with the A1111 patching. So well some hours ago I added your repo into the readme for stability as well, so more users may be moving there! |
|
I was trying to add Flux support via nunchaku in the nunchaku branch. Got to the point that the model actually "loads" and the UI launches. But since the UI asks for quite many functions from the model (e.g. getting the token count), and these functions don't exist in ComfyUI, I would have to reference the new codes and old codes and write them myself. In the end, I gave up... |
|
Found the same thing, and well if you fix the prompt parser, then a obscure reference to textual embedding happens. And then after that, I still failed to do inference lol. Not sure how to do it haha... I did also make a new branch (newmain_newforge) also based on latest forge2. Maybe adding some models on that backend is more feasible than the older one on ldm_patched. |
Out of curiosity, is it usable? PS: Would love to see you work together, specially because maintaining these kind of projects must be specially tiring when going about it solo. |
Yes, |
|
Maybe we could take a look at this project. The author seems to have put in a lot of effort, and I look forward to having the opportunity to collaborate. https://github.com/DenOfEquity/ersatzForge.git |
|
@DenOfEquity amazing, many thanks! IMO your forge2 fork bring forge2 to another and really good level. I think I will try to add changes then based on your fork instead of forge2 base fork, as it is highly factible that will bring a better experience. Also I will try to port the samplers that are on reforge which are a ton but I know that there are users that use many of them, even the more unknown ones) |
|
Oh and also, you know how feasible would to be to add WAN 2.2 support? I have some ideas but not sure how to manage it with the high and low noise models. |
|
Note that some of my changes are opinionated: 'compact' UI layout only, UI change to img2img subtabs, using The 'neg' branch of my extra schedulers extension has more samplers and schedulers, it would likely be possible to copy from there. I didn't include them directly as then I'd have another branch to WAN2.2: possibly through Refiner. Either manually by user, or detect use of WAN model and automatically setup refiner parameters. Alternatively, say that video models (even if used for 1 frame only) are outside the intended scope of Txt2img and implement it using diffusers in its own tab, or in Spaces. Using |
|
Also, the builtin extensions in my repo are either new or modified. If unchanged by me, they are not included, so some might need copying over. Exception is the legacy preprocessors, which are almost entirely replaced so if simply copied there will be duplicates in the controlnet preprocessor list. Segmentation models are the main absence, because the implementations were messy IMO and I haven't been motivated to tidy them; but I also left out a couple of depth models because the results were not up to modern standards. |
|
Perfect, many thanks for all the info and great work! And I can just imagine, you have multiple repos and is hard to maintain lol. I have first to relearn the forge2 backend mostly (I have learnt a lot in the past days but not enough yet). I would like to keep discussing a bit later, if you don't mind, via Discord (if your user is the same one as past year when we were speaking with Illya). If not, no problem! |
|
Hello guys, hope you're good. I will, maybe, now focus on newforge_newmain branch, which is based forge2 (gradio4, flux, etc) rather than the branches that are based on comfy backend (main, dev, dev2, experimental, gradio4-main). This is because after 1 week trying to implement new models, I think these branches have no really a way to do it without breaking other models and/or other features. So, preliminary, those branches (main, dev, dev2, experimental, gradio4-main) will have only bug fixes and not new features, while for that branch based on forge2, I will try first to port some things (like CFG++ samplers, schedulers, etc) and then some models, which looks way more feasible. For now it has added python 3.12 support, sageattention 2, sageattention 3 and flash attention. If someone is interested in forge2 but based on gradio3, I may give it a go but not sure how factible it is. Hope I explained myself correctly and sorry as well, as I think latest weeks some changes have had some issues when updating comfy to upstream. If you guys have too much issues, I can return main branch as how it was on April of this year. Any question is welcome, hope you have a good day. |
|
How do I check into Reforge 2/newmain_newforge (the CMD command to do it)? Also does Reforge2 still support SDXL or SD1.5? |
|
@Mr-ex777 You can do it, on the reForge folder, Sorry for the delay, I have gotten a new work and haven't managed much time to use the PC. |
I get error: Your local changes to the following files would be overwritten by checkout: when trying to swap branches, what do I do? |
|
@Mr-ex777 you can quickly try to rename the webui-user.bat file and then git checkout, and then restore that file if you used it, or keep as it is if not. |
|
NVM, it seems like I got it through git checkout -f. |
|
Welcome back, Panchovix! Thank you for your hard work and for keeping us updated. |
|
I will give a look when I have a bit more of time. I was planning to maybe do a variant with gradio 3 and latest forge2/reforge2, but have to reallly look how feasible is that. |
|
So which branch, newforge_dendev or newmain_newforge, will get the latest updates? And which one should I switch to if I want to experiment? |
|
I will start first pushing some updates on newmain_newforge as I'm learning a bit more of it's backend on the past days, and I have pushed some new commits with samplers, python3.12 support, sage 2/3, flash attention support. I have to study a bit more of the newforge_dendev backend and it's changes to start to apply updates there. Checking it quickly yesterday it has some major changes. Both are a bit experimental, but maybe dendev will be more experimental. |
|
Hi, Pancho! Is it possible to transfer "separate Text Encoder loading" to newmain_newforge branch? Also do you think --persistent-patches from forge-classic would work on your branches? Haoming02#64 Haoming02#116 and lora speedup? Haoming02#120 |
|
I thought forge2 already let you use a different TE, but maybe not? Will take a look, still have to understand a big more of the backend. I think those lora improvements would apply on the branches based on the comfy backend (aka main, dev, etc). Not sure if they're applicable for the forge2 backend. Possibly yes but implemented in a different way. |
|
Does ReForge support the Qwen model? |
|
Sadly not yet! I haven't tinkered yet to see how to add newer models. |
|
Will there be keyboard shortcuts for the gradio 4 version? The canvas in regular forge doesn't support resizing the brush with anything but the slider, which is pretty tedious |
|
I will take a look when I have a bit more time and understanding of the changes of gradio itself. |
|
Thank you for all your work. |
|
U are amazing :)!!!!! |
|
are the two branches getting merged anytime? seems frustrating having to maintain two branches. |
|
Not soon, as they are very different. |
|
Sorry for being away guys, but I think I can't manage the time to develop on the repo as I want, I'm in a bad place (mentally) so I will un-archive it as people that want to do discussions and PRs, I will still take a look. For now newmain_newforge at least has CFG++ and the samplers of reForge, which can be used to play. Again, I'm very sorry for this and raising so much expectations, I tried not to. |
|
You have my support, what you’re doing with ReForge2 really matters for the community, it’s important to have up to date UI options and not let everything funnel into one single dominant UI. A lot of people in the local diffusion community respect your work and have given good feedback, so know your efforts haven’t gone unnoticed. |
|
Do not worry lads. We always support you. You are doing great work for this project. To be honest I really enjoy your work. Still the best WebUI. |
and others
Uh oh!
There was an error while loading. Please reload this page.
Hi there guys, long time no see.
I have, for now, decided to retake reForge a little, first trying to upstream the changes of the last months up to upstream.
Then, I will try, without promises to be fair, add some features.
At the same time I will try to tackle some bugs if I can.
This probably will take a while, and maybe pushing the changes mostly to the dev branch. Just don't get too hopeful as I have to refresh a bit of my coding.
Thanks!
All reactions