FUTURE FLUX CONTENT 🤩
Added 2024-08-16 12:18:05 +0000 UTC
Alright so help me out here. Flux has been out for 2 weeks now and THERE ARE SO MANY STUFF TO TALK ABOUT OMG I don't know where to begin... I'm kinda overwhelmed...We are eating so good right now!
So..... I can either make 1 huge video talking about everything like all best practices, what model to use for which GPU, all the different must have workflows, running it on Forge, ect
but it might be a little messy with all these different parts and links and installers and stuff inside 1 video.
Or I can separate it in 2 smaller videos to make it more bearable and easier to watch, it's up you, let me know what you prefer.
Comments
Looking forward to the latest and greatest from K! Hoping it covers off Forge, Fluxgym etc. You know the simple stuff that doesn't require looking at a comfy spaghetti mess.
James Stewart
2024-09-20 08:58:39 +0000 UTCcivitai online training flux,rapid mode is very good.please teach us that
Tangbohu
2024-09-17 07:23:56 +0000 UTCyup, the limiting factor for adoption is always the worry if all the complication isn't going to pay off for something I want to generate. especially since there will be less specialized models and loras.
Faren
2024-09-15 19:19:42 +0000 UTCcan we use api to create a model which turns our life into a live rpg?
Vishwajit Pathare
2024-09-13 05:01:19 +0000 UTCbreak that flux censorship for good and we can be friends.
Federico Bicocchi
2024-09-10 20:36:40 +0000 UTCAlmost done 😀 should come out in a few days
Aitrepreneur
2024-09-10 19:35:06 +0000 UTCAny news?
StickHatzy Art
2024-09-10 19:33:53 +0000 UTCHas anything new come out from Imagen Motion/imagen to video?
Rey
2024-08-31 18:18:19 +0000 UTCI've installed that but haven't tried it out, yet. It's on my list of things to learn. lol
Christopher Moonlight Productions
2024-08-29 00:45:21 +0000 UTCIf there's a good way to do Lora training for Flux, the JoyCaption Pre-Alpha demo online seems quite impressive at generating paragraph length captions for an image. So a local version of that that could be used for Flux might be good if each caption could be manually edited like you can in Kohya for other captions, and also give it Pre/Post captions. JoyCaption Pre Alpha online demo still produces errors though but it's still detailed and impressive, if it works okay for Flux (and doesn't produce too long text). I think they tried editing Kohya for Flux but it isn't released and someone said it wasn't giving good results yet. I saw code in it (pre-release?) for SD3 training too I think. Or if OneTrainer is better a video could be shown about that. There could be comparisons of doing training for SDXL vs Flux (a particular version of it) and how the results compare, and how the file sizes compare and time taken/cost.
cool1
2024-08-26 20:00:32 +0000 UTCbrother we need the kohya tutorial!
Peruvian Dota
2024-08-24 00:26:51 +0000 UTCI'll probably stick with SDXL for a while to see how Flux evolves, but seeing everything that's going on is great o.o
Ringo
2024-08-21 22:21:22 +0000 UTCCan you share the best workflow for ComfyUI because this is tough to understand. If there was a way to use flux and forge in runpod then things would be much smoother.
Seeker Comics
2024-08-21 14:47:58 +0000 UTCPlease make a Flux Runpod 1-click installer for Forge because the Comfyui learning curb is wild. Thank you.
Seeker Comics
2024-08-21 14:35:53 +0000 UTCYES MAKE MORE PLEASE
Ulti Tor
2024-08-19 14:01:16 +0000 UTCHelp some of us add image to image, and inpainting to refine our fancy new flux image gen
Sam Shennan
2024-08-19 11:17:37 +0000 UTCwow the NF4 version of FLUX is FAST !
Sam Shennan
2024-08-19 11:17:13 +0000 UTCI would love to learn how to finetune this locally.
Peruvian Dota
2024-08-19 02:37:41 +0000 UTCI agree with this! ComfyUI is not very....well, "comfortable" to learn. I just rely on importing other people's workflows and just hope they work, since I haven't a CLUE how to start adding in things on my own. I've already used Fooocus and then Automatic1111, and (most recently) Forge. With ComfyUI that now makes four possible web interfaces, and it's by far the most confusing.
Ken L. Black
2024-08-18 14:41:01 +0000 UTCPersonally, I have found that some of the LoRAs introduced don't run on Forge as well as they do on ComfyUI....not sure why that is. But just an FYI in case you've encountered anything similar. (I voted for two separate videos. As others have said, that makes it easier to digest, and probably easier for you to edit, create, etc.
Ken L. Black
2024-08-18 14:39:33 +0000 UTCThanks for this. I'm still learning so I appreciate any and all information.
Christopher Moonlight Productions
2024-08-17 22:24:34 +0000 UTCI believe splitting the content into two videos would be more effective, considering the short attention spans most people have today. For those like me who are interested in both aspects, the second video could provide more detailed information on optimizing for different systems. Offering the content in bite-sized chunks would likely increase overall view counts, as it caters to a wider audience by keeping the information concise and focused. Plus, I really enjoy your content; I want to see you continuing to be awesome and reach even more viewers lol
Charles Hopper
2024-08-17 14:19:54 +0000 UTCBro make one adding NEGATIVE to any Workflow properly, I need to have negative STAT!
GomezBro
2024-08-17 02:06:43 +0000 UTCIt's easy... do the big steps.... (1) How to come from Automatic1111 to Comfy UI (this vid doesn't exist - most people have to copy something and don't understand how to change it). (a) basic positive/negative prompts to image (neg is not implemented in many places right now), (b) basic img2img, or sketch. (2) [next step], mix in another model? (standardize sizes, etc), or LORAs, or ? (3) [FLUX specific parameters and tricks]
Jim Gale
2024-08-16 22:04:04 +0000 UTCThere's different flux models. According to perplexity.ai: Flux-Dev - is non-commercial, meaning it cannot be used for any activities that generate direct or indirect revenue Flux-Schnell: This model is designed for faster performance, It is suitable for commercial use, unlike the Dev model. [though the text on this might not be as good as the dev one] Flux-Pro: This version is accessible only through an API [I assume that might be commercial use - but you could check but you'd probably need to pay to access it through the API]
cool1
2024-08-16 22:03:59 +0000 UTCThe flux developers have said on their site that they're working on video (probably text to video or image to video)
cool1
2024-08-16 21:58:15 +0000 UTCI would love to learn how to train it. Supposedly, someone learned to train it with a 3090. Supposedly they said that 24GB is the minimum for training but I wonder if I can get by with my 4080.
Diggy Dre
2024-08-16 20:28:51 +0000 UTCI'm a bit confused with FLUX. It makes better images, but I'm still having a blast with the oddities in SD using AnimatedDiff - Will FLUX have options like that?
patreon@winberry.com
2024-08-16 19:05:22 +0000 UTCI would like some already set up workflows with inpaiting img2img stuff like that. I have flux and comfy running now I'm just so new to comfy I need the basics.
Virtamouse
2024-08-16 18:50:17 +0000 UTCSeparate vids so you can have time to learn things in-between and end up making 3 vids to cover new goodies. So far most interesting thing I've seen is someone make a custom model by merging loras into the main checkpoint, been meaning to learn that. I'm not liking the Flux license, if it can't be worked around, I'll probably stick to things I have more custom control over.
Xodroc
2024-08-16 16:51:14 +0000 UTCAh, thank you for the clarification. Some people on YouTube act as if you can't make money on anything you create.
Christopher Moonlight Productions
2024-08-16 16:46:56 +0000 UTCEither is fine with me, but personally when I do similar things it tends to take way more effort breaking up related tasks with minimal gain. Both options have their ups and downs. Choose whichever works for you the best 😊
Ameshin
2024-08-16 16:11:43 +0000 UTCI have a 3090 so it can be done locally. I did use fal.ai Lora training online and it came out decent, but expensive ($10 for 2000 steps)
Natale Fin
2024-08-16 16:06:50 +0000 UTCI
patreon@winberry.com
2024-08-16 15:56:07 +0000 UTCThanks. Have dm'd a few screenshots.
cool1
2024-08-16 15:36:28 +0000 UTCI voted for 1 single big video to keep the overall library of information manageable on your site over the long run, but even I admit digesting the information into two smaller videos helps in the short term for me as the viewer.
Eyes_Wide_Shut
2024-08-16 15:28:32 +0000 UTCAny chance you could provide some tips on merging Flux with other models in your breakdown. Thanks for the content man!
Ian O'Mahony
2024-08-16 15:25:45 +0000 UTCNeed to get going with a Flux/Pony hybrid. When are we going to see Flux Checkpoint models? I imagine there are tons of existing image sets that are ready to be re-run through Flux generator.
Robert Waters
2024-08-16 14:54:15 +0000 UTCI started using Replicate to train Flux LoRas and it could not be any easier. Costs about $2/LoRa but nothing like having an A100 (80GB) machine cranking it out.
Robert Waters
2024-08-16 14:48:09 +0000 UTCYou don't need comfyui anymore if you don't want to :)
Aitrepreneur
2024-08-16 14:43:05 +0000 UTCNon commercial use means that you can't use the model itself to make money, like you can't sell the model itself or make people pay you to use that model. It doesn't mean that you can't use the images that you create from the model. No one would be able to enforce that anyway even if they wanted to.
Aitrepreneur
2024-08-16 14:42:33 +0000 UTCdm me some screenshots there is something very wrong here...
Aitrepreneur
2024-08-16 14:39:17 +0000 UTCSo many loras popping up. Like one every 15 mins. I am really glad that it has reenergized Forge. Comfi just isn't my cupotea. Going to keep you busy making new vids just to keep up.
Robert Waters
2024-08-16 14:22:15 +0000 UTCI tried renaming certain folders and files that might have been conflicting and done the some update command that perplexity.ai said then doing the update all bat files. then it shows more of the stuff in the options like the new vae name then trying "queue prompt" it says: Error occurred when executing UNETLoader: Error while deserializing header: MetadataIncompleteBuffer File "C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI\execution.py", line 316, in execute output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI\execution.py", line 191, in get_output_data return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI\execution.py", line 168, in _map_node_over_list process_inputs(input_dict, i) File "C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI\execution.py", line 157, in process_inputs results.append(getattr(obj, func)(**inputs)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI\nodes.py", line 863, in load_unet model = comfy.sd.load_diffusion_model(unet_path, model_options=model_options) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 647, in load_diffusion_model sd = comfy.utils.load_torch_file(unet_path) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 34, in load_torch_file sd = safetensors.torch.load_file(ckpt, device=device.type) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\safetensors\torch.py", line 313, in load_file with safe_open(filename, framework="pt", device=device) as f: (note the other stuff about the image name was because I'd moved the image2image workflow instead of the other one. This is what it says with the normal workflow (01-FLUX-FAST-LOWVRAM.json I think).
cool1
2024-08-16 14:05:23 +0000 UTCI'm so new at all of this, I need baby steps. My understanding is that as of now Flux isn't for commercial use, though, so for the moment I'm going to avoid using it as I'm working on an animated movie that I intend to monetize.
Christopher Moonlight Productions
2024-08-16 13:31:18 +0000 UTCafter clicking on "update all" in ComfyUI manager: cmd prompt says: Update ComfyUI Cmd('git') failed due to: exit code(1) cmdline: git pull -v -- origin stderr: 'error: The following untracked working tree files would be overwritten by merge:' CUSTOM NODE PULL: Fail Install: pip packages UI then says "ComfyUI has been successfully updated.". click on "close". then "restart". and "ok" to "reboot the server?" The cmd prompt then says: Restarting... [Legacy Mode] [START] Security scan [DONE] Security scan ## ComfyUI-Manager: installing dependencies done. ** ComfyUI startup time: 2024-08-16 14:21:25.676695 ** Platform: Windows ** Python version: 3.11.8 (tags/v3.11.8:db85d51, Feb 6 2024, 22:03:32) [MSC v.1937 64 bit (AMD64)] ** Python executable: C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\python.exe ** ComfyUI Path: C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI ** Log path: C:\progs\ComfyUI\ComfyUI_windows_portable\comfyui.log ####################################################################### [ComfyUI-Manager] Starting dependency installation/(de)activation for the extension [SKIP] Downgrading pip package isn't allowed: transformers (cur=4.30.2) [SKIP] Downgrading pip package isn't allowed: safetensors (cur=0.4.4) ## ComfyUI-Manager: EXECUTE => ['C:\\progs\\ComfyUI\\ComfyUI_windows_portable\\python_embeded\\python.exe', '-m', 'pip', 'install', 'pyyaml'] ## Execute install/(de)activation script for 'C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI' [!] WARNING: Error parsing dependencies of torchsde: .* suffix can only be used with `==` or `!=` operators [!] numpy (>=1.19.*) ; python_version >= "3.7" [!] ~~~~~~~^ [!] [!] [notice] A new release of pip is available: 24.1.2 -> 24.2 [!] [notice] To update, run: C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\python.exe -m pip install --upgrade pip ## ComfyUI-Manager: EXECUTE => ['C:\\progs\\ComfyUI\\ComfyUI_windows_portable\\python_embeded\\python.exe', '-m', 'pip', 'install', 'Pillow'] ## Execute install/(de)activation script for 'C:\progs\ComfyUI\ComfyUI_windows_portable\ComfyUI' [!] WARNING: Error parsing dependencies of torchsde: .* suffix can only be used with `==` or `!=` operators [!] numpy (>=1.19.*) ; python_version >= "3.7" [!] ~~~~~~~^ [!] [!] [notice] A new release of pip is available: 24.1.2 -> 24.2 [!] [notice] To update, run: C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\python.exe -m pip install --upgrade pip [SKIP] Downgrading pip package isn't allowed: kornia (cur=0.7.3) [ComfyUI-Manager] Startup script completed. ####################################################################### It also mentions stuff about "Adding extra search path checkpoints" so the same model folder from stable diffusion webui can be found and used. --- It reloads the UI. The UI already has the flux1-dev-fp8 in the unet loader section. clicking queue prompt gives: "Prompt outputs failed validation VAELoader: - Value not in list: vae_name: 'ae.sft' not in ['taesd', 'taesdxl', 'taesd3'] DualCLIPLoader: - Value not in list: type: 'flux' not in ['sdxl', 'sd3'] LoadImage: - Custom validation failed for node: image - Invalid image file: Screenshot_1.png Queue size: 0 Extra options ".
cool1
2024-08-16 13:27:50 +0000 UTCI can see one video being a "Beginner's Guide to Flux" and the other being more of an advanced guide maybe. That way you can separate out the content between them and people know roughly what to expect.
MyBrotherInChrist
2024-08-16 13:01:41 +0000 UTCyou need to update comfyui from inside the manager, click on the manager button then click update all, then it will update and ask you to restart the ui
Aitrepreneur
2024-08-16 13:00:33 +0000 UTCI've tried updating it with the 2 .bat files and through the comfyUI menu. Clicking on the update and updae python dependies gives: stashing current changes nothing to stash creating backup branch: backup_branch_2024-08-16_13_52_07 checking out master branch pulling latest changes Traceback (most recent call last): File "C:\progs\ComfyUI\ComfyUI_windows_portable\update\update.py", line 69, in pull(repo) File "C:\progs\ComfyUI\ComfyUI_windows_portable\update\update.py", line 19, in pull repo.checkout_tree(repo.get(remote_master_id)) File "C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\pygit2\repository.py", line 286, in checkout_tree payload.check_error(err) File "C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\pygit2\callbacks.py", line 99, in check_error check_error(error_code) File "C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\pygit2\errors.py", line 66, in check_error raise GitError(message) _pygit2.GitError: 42 conflicts prevent checkout - This will try to update pytorch and all python dependencies. - then says stuff like: WARNING: The script tqdm.exe is installed in 'C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\Scripts' which is not on PATH. WARNING: The scripts convert-caffe2-to-onnx.exe, convert-onnx-to-caffe2.exe and torchrun.exe are installed in 'C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\Scripts' which is not on PATH. 24.2 [notice] To update, run: C:\progs\ComfyUI\ComfyUI_windows_portable\python_embeded\python.exe -m pip install --upgrade pip I don't know if I need to edit paths for it to work. but when it's done all of the above (I've not changed path info), and I click on the .bat file to run run nvidia_gpu-LOWVRAM.bat and it's already got the flux dev workflow in there and I click on the "queue prompt" option I still get the error about: "Value not in list: vae_name: 'ae.sft' not in ['taesd', 'taesdxl', 'taesd3'] DualCLIPLoader: - Value not in list: type: 'flux' not in ['sdxl', 'sd3']"
cool1
2024-08-16 12:56:22 +0000 UTCagain it's because you need to update the comfyui, look at the patreon post and load the correct 01 json file at least if indeed all the files are correctly inside the right folders
Aitrepreneur
2024-08-16 12:44:21 +0000 UTCNo worries I'll talk about it with the new models
Aitrepreneur
2024-08-16 12:43:02 +0000 UTCHmm..I don't know I get normal images https://imgur.com/a/TKpi9tp
Aitrepreneur
2024-08-16 12:42:17 +0000 UTCI still can't get comfyUI to work with Flux. The models are downloaded but I still get the: "Prompt outputs failed validation VAELoader: - Value not in list: vae_name: 'ae.sft' not in ['taesd', 'taesdxl', 'taesd3'] DualCLIPLoader: - Value not in list: type: 'flux' not in ['sdxl', 'sd3']" error despite trying to update it. If there's some conflict in some directory that I could do something about (eg. deleting/renaming a certain directory so it gets the latest) that would help (it did seem to mention some conflict when clicking on update_comfyui.bat or update_comfyui_and_python_dependencies.bat). Or maybe if the installation installed the latest instead of installing an older version then trying to update it that might help. It might also help if the installation .bat files only installed things (including the models) that aren't already installed correctly) so it didn't take ages trying to download multiple GB files that are already downloaded (maybe it could check the file size or a checksum to check if the full file is already installed so can skip it). Also it might help if installation .bat files said on screen what files are being downloaded so it will help if there's an error part way through (people would see what file it failed to download). I voted for 2 videos since it might make things easier and more specific and any questions about them could be more specific. eg. one could talk about training Loras on it if that's becoming possible now. Maybe a video about creating videos too (there's a stable video diffusion model that can be used in comfyui but I don't know what needs to be done to use it). It seems like the developer version is better at text than the quick version (as I tried the online demos when it still allowed that for free). Maybe talk about the differences in them, including difference in licensing.
cool1
2024-08-16 12:34:52 +0000 UTCThe most utterly beautiful female elf ever with extremely long black hair dressed in luxurious elven gown.... etc, depending on the scene...
druna C.
2024-08-16 12:34:21 +0000 UTCwhat's your prompt?
Aitrepreneur
2024-08-16 12:27:34 +0000 UTCFor me by the way, I'm not sure about the workflow / model to use as I love the quality from the full dev version but it is kind of slow (on my 4080). Would like to know how to get close to that quality with the proper workflow but be able to generate much faster or perhaps have a workflow to quickly get a preview in low quality and then render the full quality. Also interested upscaling and perhaps something similar to a-detailer or inpainting. In addition I would like to know how we can train Lora models for these checkpoints. Can't find much info on that online yet. Perhaps a bit early still.
Fantastic Hank
2024-08-16 12:27:10 +0000 UTCof course
Aitrepreneur
2024-08-16 12:26:27 +0000 UTCO.k., I have a problem with Flux... It's generating only fat Ladies... It's hilarious I know, but I'm illustrating my book and it's kinda odd...
druna C.
2024-08-16 12:26:25 +0000 UTCAny plans for a 1-click installer trainer for Flux ? Or are waiting for Kohya support?
Natale Fin
2024-08-16 12:25:58 +0000 UTC
