What it does:
- Loads FLUX 2 models (DEV, Klein 4B, or Klein 9B)
- Generates 8 camera angles automatically (close-up, wide-angle, aerial, 45° rotations, etc.)
- Uses my Simple Prompt Batcher node so everything runs in one go without reloading models
The results: (see attached images)
- Klein 4B version - good quality, faster
- Klein 9B version - better detail and consistency
Built this custom node for batching prompts, saves a ton of time since models stay loaded between generations. About 50% faster than queuing individually.
Links:
Simple Prompt Batcher Node: https://github.com/ai-joe-git/ComfyUI-Simple-Prompt-Batcher
Models used:
- FLUX.2-dev: https://huggingface.co/black-forest-labs/FLUX.2-dev-NVFP4/tree/main
- LoRAs: https://huggingface.co/Comfy-Org/flux2-dev/tree/main/split_files/loras
- Klein 4B: https://huggingface.co/black-forest-labs/FLUX.2-klein-4B/tree/main
- Klein 4B splits: https://huggingface.co/Comfy-Org/flux2-klein-4B/tree/main/split_files
- Klein 9B: https://huggingface.co/black-forest-labs/FLUX.2-klein-9B/tree/main
- Klein 9B splits: https://huggingface.co/Comfy-Org/flux2-klein-9B/tree/main/split_files
Hope someone finds this useful. The batcher node works with any text input so you can use it for style variations, scene changes, whatever you need to batch test.
Description
FAQ
Comments (14)
With your batcher node my generations take way less and my pc doesnt lag?
How is that even possible? With my potato laptop this is way better.
Why? What is going on? o.O
Hello author, that prompt node is great, can it also be used on WAN SVI, after I connect it, it seems that there will be a splicing misunderstanding, a little regrettable, WAN multi-segment connection still has to be loaded and uninstalled every time
Error :
mat1 and mat2 shapes cannot be multiplied (512x2560 and 7680x3072) any solution?
Qwen 3 4b (for klein 4b) or 8b (for klein 9b)
I have a question about node 65c22b29-59aa-496b-89c6-55a603658670?
?
You are a hero- IT works so good- saving time - node - Great work!
not work( 50 series
i have a 5090 mobile, what does it do for you?
Thank you very much, this workflow very good quality for me. My hero
Thanks for this really useful workflow. It works a treat. My only comments (not criticisms!) would be the following:
I found that the turning prompts (45 and 90 degree turns) come out a bit too similar. I've adjusted the prompt for a 90 degree turn to the following "Rotate the camera 90 degrees to the left. side view. seen in profile." This seems to force a stronger side view.
The following are fun and possibly useful prompts for people to add: "Make them face away from the viewer. We see the back of their head" and "Make them face away from the viewer. They are looking over their shoulder at the viewer" which result in two noticeably different rear angle shots.
Finally, it's worth remembering that you can add a lora loader into the subnode if you need to maintain a particular look that is being lost when doing the multi-angle process e.g. a visual style, a clothing material or the face of the person.
not working for me always safe same photo and position
I keep getting the same image, am I missing something? Also the download workflow is broken the workflow in your desacription has no errors on run
似乎是新版COMFYUI導致子圖內的節點Reference Conditioning跳脫，把正負條件、圖像、VAE正確接回，即可正常運作