MengTo reports 800+ landing-page revisions with SenseNova at no model cost

• Three 3D landing pages went through 800-plus revisions in 24 hours, with a $0 model bill, according to MengTo.
• Browser feedback drove the rebuilds: MengTo’s loop captured five scroll positions, fed back console errors and discarded worse versions.
• More explicit scene instructions produced a better lighthouse in three rounds, versus 14 for an earlier attempt, in MengTo’s prompt comparison.

Two designs used reference images generated from their own screenshots as targets for later revisions. Their textures and scans came from free CC0 assets.

Three landing pages from MengTo’s 24-hour iteration run

MengTo’s SenseNova 6.8 Flash workflow repeated four steps:

1. Write the page. Flash generates the whole page as one HTML file using Three.js.
2. Render and inspect. A browser screenshots five scroll positions and collects console errors.
3. Critique and rebuild. The model scores the screenshots, lists problems and rewrites the page.
4. Reject regressions. Errors go back into the loop, and a worse round gets thrown away.

The lighthouse and ice archive added a clever extra step between rendering and revision:

• SenseNova U1.5 turned the page’s screenshots into pictures of what each chapter should look like.
• SenseNova 6.8 Flash compared subsequent screenshots against those pictures and built toward them.

MengTo described this additional image-generation step for those two designs, rather than for all three pages.

The first lighthouse prompt started from a black screen and took 14 rounds, MengTo said. A sharper prompt produced what he judged a better page in three.

The revised prompt specified how to build four parts of the scene:

• Sky
• Waves
• Cliff
• Beam

That comparison gives the human art direction a concrete role in a run where the model wrote the code.

MengTo reported more than 1,000 API calls and over 17 million tokens, with no code written by him. His $0 figure referred to the model bill for this run.

SenseNova’s official Token Plan FAQ describes the access as a free trial, with these limits:

• 1,500 calls per five-hour sliding window. Exceeding the allowance triggers throttling.
• Temporary quota exhaustion. The endpoint can become inactive until the quota window resets.
• Peak-time queuing. High concurrency can cause latency or no response.

For international access, the official installation guide lists https://token.sensenova.ai/v1 and model ID sensenova-6.8-flash-lite. MengTo also shared the prompts and loop alongside his usage totals.