I used to use A1111 extension called prompt-travel. I did not find anything like that for ComfyUI so decided to make it myself.
This workflow allows for probably any image generation model since you can just toggle cfg and negative prompt on and off. Some models do not work without cfg and negative prompt while others can or do work without. Turning it on doubles time however.
Settings:
on top below the giant wall of text you have the CFG settings and to the left of it settings to video framerate
Then there is positive and negative prompt. I decided that 4 is a good count there. More could be possible but I decided it to be unneeded. It would be possible to even make procedural but even now I have a headache thinking just thinking about the math
Next is timing setting below that. It is the meat of this. It allows 8 different curves and 4 time compression settings. From off to pow(5).
Below that is image count. Self explanatery. It is set to 300 which can take a LOOONG time. Over an hour on a 5090. There should be ETA in console.
To the left of it is Adaptive module settings. For the first time I recommand keeping it off. Turning it on requires at least 41 images. (did not test with less then 60 images)
Below you can set seed for each point. Be carefull some curves just do not work with noise and will result in grey.
Also on top you can select model and even exchange nodes when using model requiring more CLIP models or exchange it entirly for a checkpoint loader when you want to use SDXL or simillar.
The right side:
Be carefull. The right side can be laggy after generationg. On top you have your video output, on the bottom you have all imades it generates (the source of lag) and in the middle you have a graph which shows you the density of output. Does not show exactly reality when adaptive stages are off becouse the circullar corners.
Description
The base model is selected just becouse I used it last. Tested on everything from SD1.5 to Flux.2 D
Comments (8)
HI, you ssems familiar with A1111 and now using ComfyUI. Can you recomend me a ¨xyz plot¨ function that works similar to one from A1111. I mean the one where you can put different values and the output would be a picture with all pictures. in the plot.
I know what you mean. You can use images-grid-comfy-plugin. It is relativly easy to use and stateless (everything is done with 1 execution) You will however need to copy parts of workflow multiple times if you want for example different lora weights.
If you want different seed, conditioning or CFG (which would require some math) without duplicating anything you can use more_math scripting capabilities as is used in this workflow. (Sometimes is pain to use however and does not allow rendering text in images nativly).
Scratch that. You can use these 2 together. more_math to generate outputs and images-grid-comfy-plugin to put them together.
Please reply if you need anything more.
@mkDaniel Thank you for your response. Still feel like a pain for me. But I will try it. :) (I needed for the models and loras comparison) Easiest way would be to have just the workflow I would take it and put into mine. By any chance you do not have this kind of the workflow with it in?
@HugMeIntoFace Yeah, It is a pain. I just copy part which generates image like 20 times, for each changing Lora while having input seeds and prompts connected to one. Then aggregading output tensors, running through VAE and then using the grid node.
The setting of loras is the painfull process. I keep forgetting which ones I had selected and which ones I did not select. I wish I knew better way...
I tried probably everything. The most promising comfyui-essentials only generated one image instead of grid. I do not want to make another tower of samplers.
I will create 2 more nodes into more_math. One which takes folder of loras and applies each, resulting in batch of models. The other takes 1 lora and applies it to model with bunch of different strengths.
I have one problem. I do not know how to name these.
@mkDaniel Thank you that you are still looking into this! Name?! You mean name for whole process or naming the files in the process?
OK for the lora I would just make a folder with loras I would want to test and the second node would made possible to change the weight (strenght) of them. But the problem is the save it to the grid if I understand correctly?
@HugMeIntoFace I mean name for the node. Sloting it instead of normal lora loader will make each lora generate a image (using batching system comfyUI has, trying to generate 4 at once is giving me errors). Also warning it eats VRAM like candy (I have 86% of VRAM (32GB) full with 26 loras (totaling nearly 9GB) and Hidream-O1 model)
After that generation or after VAE decode place rebatch node and set batch size to at least lora count.
I did not have time making the lora strength node. (No idea how to change strength of allready applied lora yet)
images-grid-comfy-plugin is easiest to use But only allows labels of one axis.
@mkDaniel It is looking that do something that A1111 had. Making so much pain to make it work. :-D I remembering doing also the 9 or 6 in a batch then about 5 different prompts. Some things was easy to in A1111. :-D But do a 4 in batch (the 2x2) viewing in the grid in ComfyUI would be great. As you said it is eating a lot of memory. I have 24GB GPU SO had some VRAM to be used. What error the batch 4 at once saying? I tried Kimi AI to do some python scripting, with more or less not to be successful, but still working on that (mix of WAN models and exctract the lora -motion or quality) Or do you also use AI on help with errors or scripting? If yes and have positive results, can you mention which? (I am not technicaly educated enough to resolve these without AI help)