ComfyUI

ComfyUI copied to clipboard

feat: support object_storage in IS_CHANGED

[]

```object_storage``` is extremely useful for determining ```IS_CHANGED``` based on a comparison between the original value and the current value.

ltdrdata

bugfix: Pass extra_data and prompt properly to IS_CHANGED.

[]

The current extra_data and prompt are not being passed as IS_CHANGED. If there are dependencies on IS_CHANGED for decision making, an exception will occur. This patch ensures that extra_data and...

ltdrdata

Please provide controlnet for SDXL in ComfyUI

[{"_id":"66b883426f41b0ced90f2699","body":"You cannot use existing SD1.5 \/ 2.1 ControlNets with this model, those need to be retrained.\r\nSDXL 1.0 supposed to come with ControlNet models once released on July 18th. Though I would not be surprised if only the main model gets available at that time and CNs will follow sometime later.","issue_id":1688839446868,"origin_id":1627696646,"user_origin_id":829428,"create_time":1688904410,"update_time":1688904949,"id":1723368258653,"updated_at":"2024-08-11T09:24:18.653000Z","created_at":"2024-08-11T09:24:18.653000Z"},{"_id":"66b883426f41b0ced90f269a","body":"Thanks a lot for reply.\r\nI will wait for July 18th.","issue_id":1688839446868,"origin_id":1627702999,"user_origin_id":1244950,"create_time":1688905955,"update_time":1688905955,"id":1723368258675,"updated_at":"2024-08-11T09:24:18.675000Z","created_at":"2024-08-11T09:24:18.675000Z"}]

Please provide controlnet for SDXL in ComfyUI Thanks

arnold408

Scale graph canvas based on DPI factor

[{"_id":"664a21282cfd0d66a8104375","body":"Noticed you have reverted a similar DPI fix previously https:\/\/github.com\/comfyanonymous\/ComfyUI\/commit\/4796e615dd7faad38429fc8e716e3a817a28c526.\r\n\r\n> Revert DPI fix since it caused more issues than it solved. \r\n\r\nI assume \"more issues\" means the canvas would still be blurry with that fix. I guess the issue is the previous patch uses `el.offsetWidth` and `el.offsetHeight`, which are rounded integers, causing scaled width & height to be inaccurate.\r\n\r\nHence in this patch, width and height are get from more accurate [`getBoundingClientRect()`](https:\/\/developer.mozilla.org\/en-US\/docs\/Web\/API\/Element\/getBoundingClientRect) in float number form. And in the last `Math.round()` are applied to scaled width & height otherwise the decimal part would be implicitly dropped after setting to canvas element.\r\n\r\nIf this still causes issues in your end, than at least a toggle should be giving to opt-in\/out DPI scaling as it's critical to HIDPI screen user.","issue_id":1688839446917,"origin_id":1627563738,"user_origin_id":24871166,"create_time":1688865444,"update_time":1688865444,"id":1716134184805,"updated_at":"2024-05-19T15:56:24.804000Z","created_at":"2024-05-19T15:56:24.804000Z"},{"_id":"664a21282cfd0d66a8104376","body":"The issue with this is that it breaks zooming and unzooming in the browser. When unzooming the canvas only takes part of the window and when zooming you no longer see the fps counter on the bottom left.","issue_id":1688839446917,"origin_id":1630105831,"user_origin_id":121283862,"create_time":1689050493,"update_time":1689050493,"id":1716134184811,"updated_at":"2024-05-19T15:56:24.810000Z","created_at":"2024-05-19T15:56:24.810000Z"},{"_id":"664a21282cfd0d66a8104377","body":"OK, I see the problem here. \r\n\r\nlitegraph.js internally uses `canvas.width` & `canvas.height` as its absolute right & bottom viewport bounds, and the fps counter is positioned to its left-bottom bound. So lets say if we scale the canvas by 2x, the internal viewport would actually be 2x2 times larger than browser viewport, thus the fps counter became invisible.\r\n\r\nThe \"unzooming\" problem can be work around by limit the minimal scale to `1`.\r\n```diff\r\n- const scale = window.devicePixelRatio;\r\n+ const scale = Math.max(window.devicePixelRatio, 1);\r\n```\r\n And there is a ad-hoc fix for \"fps counter\" positioning.\r\n```diff\r\ndiff --git a\/web\/lib\/litegraph.core.js b\/web\/lib\/litegraph.core.js\r\nindex a60848d..05f52ce 100644\r\n--- a\/web\/lib\/litegraph.core.js\r\n+++ b\/web\/lib\/litegraph.core.js\r\n@@ -8148,7 +8148,7 @@ LGraphNode.prototype.executeAction = function(action)\r\n **\/\r\n LGraphCanvas.prototype.renderInfo = function(ctx, x, y) {\r\n x = x || 10;\r\n- y = y || this.canvas.height - 80;\r\n+ y = y || this.canvas.offsetHeight - 80;\r\n \r\n ctx.save();\r\n ctx.translate(x, y);\r\n```\r\n\r\n\r\nI guess the issue should be addressed in upstream first to make litegraph.js `window.devicePixelRatio` aware: https:\/\/github.com\/jagenjo\/litegraph.js\/issues\/181\r\n\r\nBut I can push a v3 patch if it's OK for you to include the workaround above, though the viewport problem in litegraph.js would still remains.\r\n\r\nedit1: Note the `renderInfo` fix can also be implemented by overriding and wrapping the original `LGraphCanvas.prototype.renderInfo` in `app.js` to ease upgrading of `litegraph.core.js`.\r\n\r\nedit2: Pushed the updated patch anyway, it can be useful for try-out and\/or reviewing.","issue_id":1688839446917,"origin_id":1630203068,"user_origin_id":24871166,"create_time":1689056026,"update_time":1689059326,"id":1716134184814,"updated_at":"2024-05-19T15:56:24.814000Z","created_at":"2024-05-19T15:56:24.814000Z"},{"_id":"664a21282cfd0d66a8104378","body":"Well I merged it: https:\/\/github.com\/comfyanonymous\/ComfyUI\/commit\/cef30cc6b66fc6b081d3976a547f8abc56a4c7ae","issue_id":1688839446917,"origin_id":1630285072,"user_origin_id":121283862,"create_time":1689060048,"update_time":1689060048,"id":1716134184817,"updated_at":"2024-05-19T15:56:24.817000Z","created_at":"2024-05-19T15:56:24.817000Z"},{"_id":"664a21282cfd0d66a8104379","body":"Fine, I just fixed a typo in the last version: \"viewpoint\" -> \"viewport\".\r\n\r\nThough I personally would prefer to use overriding method, but it's your choice.\r\n\r\nAnyway, thanks for making progress.","issue_id":1688839446917,"origin_id":1630293420,"user_origin_id":24871166,"create_time":1689060418,"update_time":1689060418,"id":1716134184820,"updated_at":"2024-05-19T15:56:24.820000Z","created_at":"2024-05-19T15:56:24.820000Z"}]

Similar to fixes in litegraph.js editor demo: https://github.com/ernestp/litegraph.js/blob/3ef215cf11b5d38cc4f7062d6f78b478e2f02b39/editor/js/code.js#L19-L28 Fixes #161

EHfive

CoreML intergration?

[{"_id":"662f1383d8e74d79d10544d8","body":"I would like to but I don't have a mac.","issue_id":1688839446956,"origin_id":1626369312,"user_origin_id":121283862,"create_time":1688773859,"update_time":1688773859,"id":1714361219486,"updated_at":"2024-04-29T03:26:59.485000Z","created_at":"2024-04-29T03:26:59.485000Z"},{"_id":"662f1383d8e74d79d10544d9","body":"> I would like to but I don't have a mac.\n\nAny way I could like help? In like being a tester or something like that? Cuz I just attempted to to generate on the SDXL model... took 45min to make 1 image. I know Coreml model doesn't exist for SDXL yet. But it would be soon and 1.5 generation are 50% faster on coreml on Mac ","issue_id":1688839446956,"origin_id":1626370744,"user_origin_id":62298209,"create_time":1688774086,"update_time":1688774086,"id":1714361219489,"updated_at":"2024-04-29T03:26:59.488000Z","created_at":"2024-04-29T03:26:59.488000Z"},{"_id":"662f1383d8e74d79d10544da","body":"SDXL 1.0 will be officially supported on https:\/\/github.com\/apple\/ml-stable-diffusion soon, with optimized models for ANE. These can run directly from python via coremltools, so might be a good option for folks without a cuda device. Happy to help out with the integration","issue_id":1688839446956,"origin_id":1654062900,"user_origin_id":1981179,"create_time":1690478356,"update_time":1690478356,"id":1714361219493,"updated_at":"2024-04-29T03:26:59.492000Z","created_at":"2024-04-29T03:26:59.492000Z"},{"_id":"662f1383d8e74d79d10544db","body":"@comfyanonymous is there a way now in ComfyUI to select backends? If so, a coreml backend wrapper can be plugged.","issue_id":1688839446956,"origin_id":1709658906,"user_origin_id":654041,"create_time":1694073485,"update_time":1694073485,"id":1714361219496,"updated_at":"2024-04-29T03:26:59.496000Z","created_at":"2024-04-29T03:26:59.496000Z"}]

Any plans on allowing to run CoreML models on ur Ui? since Apple CoreML models run waay better than normal safetensors on mac.. much more faster

Maki9009

BUG: Only using 7.7GB Max out of 24GB Nvidia 3090 RTX

[{"_id":"66b88360d33fb4691e10b49f","body":"Can you give me some numbers? If you downloaded the latest standalone comfyui should be faster even though it's using less memory.\r\n\r\nlaunch it with: --gpu-only if you want it to use more vram it will lower total gen time but won't increase it\/s\r\n","issue_id":1688839446978,"origin_id":1626368186,"user_origin_id":121283862,"create_time":1688773695,"update_time":1688773695,"id":1723368288339,"updated_at":"2024-08-11T09:24:48.339000Z","created_at":"2024-08-11T09:24:48.339000Z"}]

I've tried increasing the batch size up to 8 or more, still only uses 7.7GB Max memory on the 3090. Automatic1111 is using the full 24GB while rendering, and is...

2zzly

[bug] Custom nodes can overwrite core nodes through class and display name mappings

[]

There's no protection or guard for core nodes in nodes.py's definition against being overwritten by custom nodes. `nodes.py`: `NODE_CLASS_MAPPINGS` | `NODE_DISPLAY_NAME_MAPPINGS` I was writing custom logic to split all custom...

dotJack

Modify context menu styles

[]

I am uncomfortable using the default context menu of litegraph, so I made a small modification to improve it. Before:  After:  Now, The button area is enlarged, which...

jnyfil

The basic_api_example.py seems to be outdated

[]

Is the format changed？I use it and the ComfyUI seems to be not working properly(the generated image is disordered) and i can't find the "prompt section" in saved .json file

daxijiu

Fixed Pull Request 749: Dynamic node loading

[]

Fixed https://github.com/comfyanonymous/ComfyUI/pull/749 which broke cli_args in its last commit. Please refer to the original PR for details.

KaruroChori