Skip to content
- Continue: make the next scene from where a MiniMax H3 video ends
- MiniMax H3: Sparse on/off switch — fast sparse attention, on by itself when the Civitai model says so
- A MiniMax H3 model from Civitai: paste its link in «Your own worker» for video with sound from your picture (MiniMax H3 Universal).
- The «Any Civitai» workers are now called Universal (SD 1.5, SDXL, FLUX.1, Chroma, Z-Image) — same repos, same links. Set up: a «Stuck here?» button on each step.
- Plus: **Bulk requests** — one prompt per line, «×3» for several of one; RunPod makes them in one batch, and they come in even if you close the page. How it works
- Pricing: the launch price lasts until December 31, 2026, and Plus includes 12 months of updates or more from the day you buy.
- Text is one size bigger.
- Manual, Setup, News and Feedback moved from the top menu to the bottom of About.
- Pick the picture’s shape: landscape, square or portrait. Make it is gone, so nothing passes through a Cobalt Pear server and the app records nothing about you. Plus is now on the Pricing page.
- About is back in the menu
- A Pricing page in the menu — Cobalt Pear is free to use; you pay RunPod only for GPU time.
- The setup steps, About and the pack fields moved to the manual, with a new Cost and speed page. Make it names the models it makes and shows which one it’s on; Make it, by hand and the Civitai token each sit in a box of their own. LoRAs built in: one block for all your models, with a Copy button for each line.
- News moved to the manual — the News link at the top opens it.
- Settings: Backup — save your models, endpoint IDs, LoRAs and settings to a file and load them back after clearing site data or in another browser (keys and tokens are left out).
- Settings: Plugins — add features from a plugin file. Only plugins signed by Cobalt Pear are taken; a changed or unknown one is refused, since a plugin can read your RunPod key.
- Any FLUX.1, Chroma, Z-Image or SD 1.5 model from Civitai (FLUX.1 and Z-Image from Hugging Face too) — not just SDXL.
- LoRAs: build them into your worker and pick which ones each picture uses, or add them by link.
- LoRAs on A1111 workers, up to 8 LoRAs per model, LoRAs built into a worker, and a Custom button on Hugging Face SDXL models.
- Add your own LoRAs by pasting a Civitai or Hugging Face link (up to 4), and see a warning when one has no effect
- News is grouped by day.
- Edit and picture-to-video templates: your picture now reaches the worker.
- Add any SDXL model from Civitai with its link, make a pack by filling in boxes, and clearer setup steps for RunPod and Civitai
- Models: each model you added shows what it is — Text To Image, Image To Video…, NSFW, Gated, or Custom for your own worker. «Use as picture» no longer changes your models; with none that takes a picture, the picture says it is left out.
- Picture: pick one you made here, right from the Picture button — and «Use as picture» under every result, even before you have a model that takes one.
- Negative: with several models chosen, leave it empty and each gets its own — type words to send the same to all, ✕ to go back. Feedback is now in the top menu.
- Chat: all models sit in one list, grouped like the Models tabs — click to tick them on or off, so the row stays short however many you have.
- Make it: pick any of 127 models, tick «Let Cobalt Pear make it», and it’s ready on your RunPod in a few minutes.
- Cobalt Pear is out. Start with Z-Image Turbo from the Models list — no GitHub, no build.