DeepSeek-R1 Enhanced Flux Prompts
Added 2025-01-31 20:29:41 +0000 UTCVideo - https://youtu.be/T7nufrDmc9A
DeepSeek. It's all the rage at the moment, so I figured I'd give it a spin in ComfyUI to see how good it was at prompt enhancing for image generation!
The following models were used to compare against:
DeepSeek-R1 (14B)
Gemma2 (9B)
Llama 3.2 (3B)
Olmo (7B)
Phi4 (14B)
The TL;DR is yes, it's very good. Is it the best? Not fully convinced yet :) I'm liking Olmo at the moment, but time will tell.
I also found the new gradient estimation sampler gave decent results with Flux, so certainly taking a look at if you've not done so as yet. It certainly seems to pair nicely with the kl_optimal scheduler.
Links
https://ollama.com/library/deepseek-r1
Workflow
Just a single workflow this time as it's simply for testing out the prompt capabilities of various large language models.
It includes various extra nodes from previous videos, such as Detail Daemon and WaveSpeed.
If you don't have ollama installed yet, you'll need to do that first, as well as download whatever models you want to test from the ollama library, E.g.
ollama pull deepseek-r1:14b
LLM mode can be turned off too, should you just want the basic prompt.
The regex node used is from comfyui-lf - https://github.com/lucafoscili/comfyui-lf (as apparently manager can't detect this one)
WaveSpeed (covered previously) - https://github.com/chengzeyi/Comfy-WaveSpeed
Derfuu - https://github.com/Derfuu/Derfuu_ComfyUI_ModdedNodes
Example screenshot of a completed workflow run:
Also included is the full-size montage image of the test results.
Comments
Thanks. I've added the links to those in the main description as manager isn't detecting them
Nerdy Rodent
2025-02-05 17:48:47 +0000 UTCI have the following errors while loading a workflow: DF_Get_latent_size ApplyFBCacheOnModel LF_RegexReplace
Ruslan Urmanov
2025-02-04 19:39:17 +0000 UTCok got it. It's there in the manager. Not sure why it didnt find the nodes before. Thanks for your help
guni
2025-02-02 17:21:19 +0000 UTCLooks like it is back in the manager search list now (though it no longer has the little wave icon in the name)
Nerdy Rodent
2025-02-02 13:55:39 +0000 UTCThe regex node is in comfyui-lf - https://github.com/lucafoscili/comfyui-lf
Nerdy Rodent
2025-02-02 13:51:02 +0000 UTCI'm guessing you downloaded the attached json file in the end? I've not heard of Anything LLM, so my first guess it something to do with that. Typically when an LLM gives rubbish answers, you can tweak the top_p and top_k
Nerdy Rodent
2025-02-02 13:50:01 +0000 UTCThat one is in comfyui-lf - https://github.com/lucafoscili/comfyui-lf
Nerdy Rodent
2025-02-02 13:47:44 +0000 UTCWhere can I get that? When I open the wf it is a missing node. Using manager doesnt list the missing node
guni
2025-02-02 12:15:05 +0000 UTCwhen I go to install missing node, couldnt get the regex node... any idea how to install the missing node?
guni
2025-02-02 05:41:46 +0000 UTCAm i missing something? I can't find the workflow. The image doesn't have the workflow builtin.
Gary
2025-02-01 20:15:18 +0000 UTCI don't know what's up, but when I use any of the deepseek models i get a bunch of jiberish answers that have nothing to do with the prompt. It goes on like a yapping dog, changing subjects and continuing to spew out answers. I'm using Ollama and AnythingLLM to test it. what's the deal?
Gary
2025-02-01 19:59:14 +0000 UTCinstalling this would help https://github.com/chengzeyi/Comfy-WaveSpeed
p tjie
2025-02-01 13:14:44 +0000 UTCI was able to resolve by manually install wavespeed via git pull, described in the βSpeed boost Fluxβ earlier episode
Sebastian Antony
2025-02-01 12:46:53 +0000 UTCgot this error too...
p tjie
2025-02-01 11:11:45 +0000 UTCThank you for creating this. Though I updated ComfyUI via Manager, I'm having an issue loading just one node: ApplyFBCacheOnModel - Is there any guidance you can provide me? Thanks
Sebastian Antony
2025-02-01 10:29:22 +0000 UTCAwesome for putting this together so quickly! Yeah, Great job!
Lord Banzai
2025-02-01 04:07:30 +0000 UTCRegex to the rescue!
Nerdy Rodent
2025-01-31 21:23:00 +0000 UTCGood job. I already added DeepSeek-R1 to my workflows, but I need to parse and remove the "thinking" part of the context... and I'm too lazy for that, so I went back to use LLama3.2 π
Alexander Garzon
2025-01-31 20:34:30 +0000 UTC
