Generate Images in Chat: Create in Chat on LU Cloud and the Local Way
In Locally Uncensored you can ask for a picture in the chat. On LU Cloud, Create in chat (Beta) lets the chat model put together an order that you approve. On your own machine, the agent renders through the local Create lanes.
Create in Chat on LU Cloud
Where the switch is
Create in chat exists in Cloud mode only. In a chat, open the Plugins menu. There is a row called Create with a Beta mark and the words "image · video · music · voice".
If your account does not have Create in chat yet, the row is not there. If it is grey, the line under it says why, for example "This model cannot call tools. Pick another one to use Create." or "Not available with this model." A context window that is too small gets its own line.
From request to order card
Ask in plain words. If the request is unclear or an input is missing, the model is told to ask first. If it is clear, it does not ask which model or setting to use. The order card is the question.
The order appears as a card, and nothing runs until you approve it. It shows:
- How many items the order has, and the state "Waiting for your OK"
- One row per item with its title, a short detail line and its price in credits. An estimate has a ≈ in front, and an item that depends on an upload says "Price after upload"
- For an item with several steps, each step with its model
- The total in credits, and what this chat has already used when there is something
- A Start button with the total on it, and Not now. You can also just reply "yes"
Under a waiting card, LU adds: "Create in chat is new. Read the order before you start it." The Beta mark next to it gives the reason: "A small chat model often gets an order wrong, so read the order before you press Start." Ordered pictures follow your account's content policy, like everything rendered in the cloud. What credits cost is on the pricing page.
While it runs, and after
After Start, each row shows how long it has been running. Stop ends the order, and the card says "Items that have not started are not charged".
A finished picture appears under the card. Since 3.0.6 every result in Create shows the prompt it was made from, pictures ordered from a chat included. A click opens the large view with the prompt and a Copy button.
3.0.6 also fixed pictures made in the running session. Using one as the source in Edit used to fail with "Could not load that gallery image: Load failed". The fix also covers the chat model being shown a picture it made.
Characters
The chat can also order the Character Studio: train a character from 4 to 30 photos, or create images with a trained one. A training run leaves no picture in the chat, because a character is not an image. In Cloud mode your characters are listed under Create, Character, "Use character". Erase needs a painted mask, so it is not available in chat.
The Local Way: Your Own Models
In Local mode there is no Create switch. You have two ways to make pictures on your own card.
In the chat, through the chat tools
The chat tools, which are on out of the box, and Agent mode both carry the tools image_generate and video_generate. They render through the Create lanes, and in Local mode that means your own image model. Image Generation is set to Ask First by default, so a dialog shows the tool and its input with Approve and Reject before anything runs. It needs a model that can call tools, see the Agent chapter.
In the Create tab
On Windows and Linux every local lane of the Create tab runs on ComfyUI, which the app installs and starts for you. On a Mac with Apple Silicon, Create makes images on the Mac itself without ComfyUI, and the Mac build is in beta. The first local image needs the Juggernaut XL starter checkpoint, about 6.5 GB. Image generation starts at 6 to 8 GB of graphics memory. The Create chapter covers every lane.
Which One to Pick
Pick the local way if your card has the memory and your pictures should stay on your machine. Pick Create in chat if you have no suitable card, or want to talk a picture into shape and see the price before it runs. In a browser, the image generator online covers the hosted image models. Locally Uncensored vs ComfyUI compares the two local setups.
Locally Uncensored is AGPL-3.0 licensed and built by PurpleDoubleD. Corrections welcome on GitHub Discussions or in the Discord.