Skip to content
Lab

commit c19e5a0 ·

Higgsfield → Runway → HeyGen → Kling: why each switch happened

A log of tools tried for the Haj, what each one did well and where it broke.


The Haj didn't start in the tool that makes him today. This is the short version of how he got there. It is a log, not a ranking: each of these tools is good at something, just not at all of what this character needs.

1. Higgsfield

The starting point, and still where most of the Haj's still images come from. For video, several models were tried inside it, including Seedance, Genjutsu and Kling. Every run had a defect somewhere: a face that drifted, motion that over-acted, or mouths that didn't follow the words.

Kept: image generation and look development.

2. Runway

Strong, cinematic motion. But the Haj is a talking head with jokes, and the trade-off between expressive motion and controlled, in-sync speech kept landing on the wrong side for this character.

3. HeyGen

The avatar system solves lip-sync, which is exactly what the earlier tests lacked. The cost was performance: the delivery felt like a presenter, not a sarcastic 70-year-old. Comedy lives in timing and tiny gestures, and those were the first things to go.

4. Kling motion control

What stuck. Feeding a real driving performance plus one reference image means the timing, gestures and mouth shapes come from a human, and the model only has to keep the Haj looking like himself.

Side note: voice

A test of reshaping the Haj's voice with a dedicated voice tool, applied to the original take, didn't hold up. It was dropped rather than forced.

Side note: content filters

Prompts that named a famous anime character to describe the straw-hat look were blocked by the copyright filter. Describing the look itself, without the name, works.

Next note05d2b7fThe HAJFLIX intro, written in code with Remotion

HAJFLIX://contact

Talk to the Haj

Pick the way you like. WhatsApp is the fastest; he answers between naps.

«عندك شغل؟ الحاج فاضي… لحدي ما يجي الزبون الجاي.»