Persistent history and media archive
Every job and result is saved in a real database and storage before the provider's 7-day links expire, instead of temporary server space.
Updated
Build log
A private AI image and video studio for Kunkun Visual. This board is public and read-only — it updates with every feature we ship.
Every job and result is saved in a real database and storage before the provider's 7-day links expire, instead of temporary server space.
The Generate button shows an estimated cost from Higgsfield, and expensive or multi-shot jobs ask for one confirmation.
Real generation is switched on only after one owner-approved test per model family.
Prompt, pick a model, ratio and quality, add references where the model supports them, and get results in the feed, in the Google Flow style with fewer knobs.
Open any result to see exactly how it was made and put the same settings back into the prompt bar in one click.
Send any image into a Kling 3.0 or Seedance 2.5 video as its start or end frame with one click or a drag.
A scene plus up to six shots, each with its own prompt and seconds, sent to Kling 3.0 as one clip with a draft-then-final option.
Turn any paused video frame into an image, or start the next shot from a clip's last frame, to build sequences longer than 15 seconds.
Describe a change to an image and get a new version alongside it, while the original stays untouched.
One archive with search, filters, collections and multi-select, where any result can become a reference or a frame for the next generation.
A short, curated set of camera moves, lenses, lighting and colour palettes, offered as one-tap chips on the model that supports them.
Extend a finished clip or change it with a prompt using Seedance 2.5, offered as plain actions on the clip.
Dark olive-and-lime brand tokens and a short list of the only places things are allowed to move, all kept under 0.4 seconds.
All 45 Higgsfield models are read from their official schemas, so the studio only shows settings a model really accepts.
A stand-in provider that plays out every outcome (done, failed, blocked, cancelled, unclear), so each screen can be tested without spending credits.
Every job shows its real status with a running timer, can be cancelled while queued, and is never re-sent automatically when its outcome is unclear.
Keep the studio private while this build log stays public.
The studio layout on the mock provider: a single docked prompt bar for image and video, a results feed, and a jobs tray.
A one-page plan for the studio plus clickable screens for image, video and multi-shot, approved before any studio code is written.
The original ComfyUI node pack now lives privately under our account, with the original kept as a read-only upstream.
A clean rebuild on Next.js 16 and React 19.2, replacing the original source we couldn't access.
This page: a read-only kanban and changelog anyone can open without signing in.
Seven researchers studied Kling Multishot, Kling video, Google Flow image and video, six peer studios, motion, and the Higgsfield API — 238 sources in total.
The original site's source wasn't available, so Render Gangsta is being rebuilt from scratch. The Higgsfield ComfyUI nodes were copied privately, and their official model catalog becomes the foundation. UX research on Kling Multishot and Google Flow is underway.
The studio design is settled on paper: one prompt bar and one feed for both image and video, multi-shot as a list of up to six shots that Kling 3.0 renders as a single clip, an estimated cost on the Generate button before anything is sent, and job states that never dress a failure up as a success. Every feature was checked against what the Higgsfield API really accepts; 14 details are flagged for a supervised test before the features that depend on them ship.