Which model gives the most convincing CLAY texture out of the box (before your post-processing)?

by @oliviamiller_films in c/claymation ·

Purely on raw clay-ness of the generation before you grade/retexture it. Which has the best native feel for plasticine?

  • Kling (with a clay LoRA/strong prompt) (43 votes)
  • Hailuo (good matte surfaces) (45 votes)
  • Midjourney (best clay STILLS, then animate) (51 votes)
  • Runway (clean but plasticky) (54 votes)
  • A dedicated clay-trained LoRA on an open model (Wan/Hunyuan) (50 votes)
  • None are good raw, it's ALL in the post (55 votes)

298 votes · poll closed

291 points · 6 comments

Comments

  • @sakuramueller · · 98 points

    voted 'it's ALL in the post' because honestly it is. the raw gen is a starting point from any of them, and the clay-ness lives in the grade (matte), the frame rate (twos), and the texture overlay (thumbprints). the model choice matters way less than what you do AFTER. a mediocre gen + great post beats a great gen + no post every time.

  • @kaikimlab · · 84 points

    this is the hard truth that beginners don't want to hear — there's no 'good clay model', there's a clay PROCESS, and the generation is like 30% of it. the people with convincing clay aren't using a secret model, they're doing the post work you're not.

  • @ariawhitelab · · 70 points

    Midjourney for STILLS then animate separately is a legit strategy for clay specifically — MJ's clay texture on a still is gorgeous, better than any video model's, and you can bring that texture quality into the animation via img2video. start from the best possible clay frame.

  • @ethansingh_studio · · 48 points

    the 'best still then animate' approach wins for clay because the texture quality of the FIRST frame propagates. garbage clay in frame 1 = garbage clay throughout. invest everything in a perfect clay still, then move it.

  • @dmitrimeyer_art · · 44 points

    real set, AI character, or full AI? the handmade warmth makes me think there's real clay somewhere in this pipeline.