One of the ongoing challenges with AI generated art has been its inability to draw hands correctly. They are often deformed or have way too many fingers which really would detract from the artwork. Unless you were going for deformed hands then you are in heaven.

Finally there has been a breakthrough in this space where a AI Model checkpoint Protogen x3.4 Official Release (ProtoGen_X3.4) is available for Stable Diffusion that can draw hands correctly. This is an amalgamation of multiple checkpoint, being merged into one. It comprises of 5% of roboDiffusion_v1.ckpt, openjourney-v2-unpruned.ckpt, analog-diffusion-1.0.ckpt and rpg_v2Beta.ckpt.

ProtoGen_X3.4 can produce natural looking hands that are suited to the composition of the image and is able to keep the length of figures correct including the number of figures. The samples and test images submitted on the discussion page already demonstrate the correctly formed hands.

As suggested on their page it is recommended to remove “ugly” from the negative prompt when creating images. Seriously, removing off (ugly) on negative prompts brings out some really detailed shots of what real life consist of, decay, rubble, grass, worned clothing…Have fun and keep it fluffy!

Below are sample images referenced from the ProtoGen_X3.4 release page.

If you are convinced then you can head over to the Protogen x3.4 Official Release page to download this CKPT file (5.57GB) or download from Huggingface. If you are using a Paperspace notebook you can upload it to your machine within a few seconds by using this technique I referenced in another post, how to quickly upload model CKPT to Paperspace.

Prompt: modelshoot style, (extremely detailed CG unity 8k wallpaper), full shot body photo of the most beautiful artwork in the world, english medieval witch, black silk vale, pale skin, black silk robe, black cat, necromancy magic, sexy, medieval era, photorealistic painting by Ed Blinkey, Atey Ghailan, Studio Ghibli, by Jeremy Mann, Greg Manchess, Antonio Moro, trending on ArtStation, trending on CGSociety, Intricate, High Detail, Sharp focus, dramatic, photorealistic painting art by midjourney and greg rutkowski

Negative prompt: canvas frame, cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), wierd colors, blurry, (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render

Steps: 30, Sampler: DPM++ SDE Karras, CFG scale: 10, Seed: 1495009790, Face restoration: CodeFormer, Size: 760×1024, Model hash: 60fe2f34, Denoising strength: 0.5, First pass size: 0x0

Testing Protogen x3.4

I uploaded the CKPT file to my own Paperspace Notebook and ran it using Automatic1111. Initially I tried with the same prompt as above but then started to vary it and have more than one person in the image. The results are better than before but not quite there yet. However, I am not being too critical about this model as it is only the first release. I’m certain that the team behind Protogen will keep improving this model to produce more consistent results.

Well formed hands
Something weird happened here
Woman’s Hand is not correctly formed
Well formed hands
Left hand with three fingers
Too many fingers in this case
Well formed hands
Well formed hands
Well formed hands

Tutorial Video

I found this wonderful tutorial video that shows how to setup and use this model in Google Colab notebook with Automatic1111. If you can get past the accent the information being shared is useful and demonstrates how to use this model.

Overall I feel with this Protogen model we are heading on the right path of improving hand formation in images created using AI and hopefully in a few months we will have this challenge completely tackled. Feel free to give this model a run, it was certainly fun for me to try various images and see the results.

If you'd like to support our site please consider buying us a Ko-fi, grab a product or subscribe.