ComfyUI: Batch Generate Text 2 Image. Easy Tutorial for automating multiple prompts
HTML-код
- Опубликовано: 27 май 2024
- Tired of making one image at a time in ComfyUI? This video will show you a super easy trick to batch process your text prompts in ComfyUI and create a whole explosion of AI-generated art!
In this video you'll learn:
- How to use ComfyUI to make tons of images at once
- Common pitfalls and errors with custom nodes used.
- Prompt guide for generating awesome text prompts to get the pictures you want.
- Combining multiple prompts to generate unique concepts
This video is perfect for beginners with basic ComfyUI skills who want to learn how to become more efficient with AI art generation.
00:00 - Background
01:08 - Install Custom Nodes
01:37 - Setting up a basic Text to image workflow
03:12 - Modify your ComfyUI workflow for batch processing
03:58 - Create batch text file for multiple prompts
04:18 - Use text file in workflow
06:20 - Errors
06:42 - Bonus - Stable Diffusion Prompt Guide
07:38 - Pitfall
08:48 - Bonus - Combining prompts
Links mentioning in the video:
Workflow: openart.ai/workflows/veDUGlU8...
Text Prompts (See description for workflow): openart.ai/workflows/veDUGlU8...
ComfyUI-Inspire-Pack: github.com/ltdrdata/ComfyUI-I...
ComfyUI-Impact-Pack: github.com/ltdrdata/ComfyUI-I...
Prompt Guide: / i_am_trying_to_teach_c...
Formatting Prompt:
I would like to bulk import these prompts into a single text file. In this file each prompt has a "positive:" and "negative:" prefix, followed by the positive and negative prompt respectively. There is a empty line separating the positive and negative prompt. Each prompt then needs to be separated by a series of hyphens ("---"). I've listed an example output using 3 random prompts below. Use the prompts from the previous response and format it to match this style.
###Example output###
positive:1girl is walking through street,
raincoat, yellow umbrella
negative:text, watermark
-----------------
positive:museum, people are looking paintings, abstract
negative:text, watermark
----
positive:battle ground of space ships
negative:text, watermark
###End of example Output###
Music by: www.bensound.com/free-music-f...
License code: PXAIHHGD4HM7YMSB - Наука
Splendid tutorial that points directly. Thank you so much for your effort.
nice, but i want the same prompt for say 20 generations, then the next prompt for 20, then the next for 20...etc. any ideas? it only says sequential in the docs so hmm.
when I modify the txt prompt it doesn't update only if I closed comfyui and open again , how to fix this ? and is there another to directly input text in the comfyui and not in a txt ?
Great tutorial, but is there a way to randomize the seed of each image generated? It seems since technically comfyui see's it all as one prompt queue, the seed stays the same for all generated images. I'm trying to make a batch img2img workflow that has 4 images for each facial expression (4 angry, 4 sad, 4 happy etc.) and all 4 angry/sad/happy faces come out looking exactly the same.
Other than that, this tutorial helped immensely, wouldn't have gotten as far as I did without it!
Yeah using different seeds for each image in the batch is a tough one. I've managed to get more variation with:
1) converting the seed for the sampler to an input, and connecting it to the rgthree seed node (set to randomize each time)
2) connecting the empty latent image to "LatentBatchSeedBehavior" node set to random seed behavior, and then feeding the output of that to the ksampler
github.com/rgthree/rgthree-comfy
docs.getsalt.ai/md/Comfy/Nodes/LatentBatchSeedBehavior/
p.s. I show how to add variation to batches in this section of my new video: ruclips.net/video/mecA9feCihs/видео.html
It is important to update ComfyUI to the latest version, because without updating the node LatentBatchSeedBehavior may be missing
❤🔥❤🔥❤🔥