SORCERESS
Browse the wiki

AI Image Gen

Updated July 4, 2026Open the tool

Open AI Image Gen

AI Image Gen is Sorceress’s main image generation workspace. Use it to create images from text prompts, compare one or more cloud models, guide supported models with reference images, generate with supported local GPU models, and manage results in a searchable personal gallery.

On desktop, the page is arranged in three areas: the model picker on the left, your gallery in the center, and prompt/generation controls on the right. On mobile, the Models and Prompt panels open from the top toolbar.

What it does

AI Image Gen lets you:

  • Generate images from a written prompt.
  • Choose one or more cloud models for side-by-side comparison.
  • Use Pro Local GPU generation with supported installed local models.
  • Upload or drag in reference images for models that support image guidance or image editing.
  • View, download, remove, and inspect reference images before generating.
  • Choose aspect ratio and quantity.
  • Save generated images to your library, or keep local GPU generations temporary when desired.
  • Search, favorite, download, delete, and browse generated images.
  • Reuse a previous generation’s prompt, model, aspect ratio, settings, and references when available.
  • Share a generation to Prompt Lexicon with an optional comment.
  • Send completed images directly into other Sorceress tools.
  • Drag completed images into WizardGenie Explorer when Image Gen is embedded in WizardGenie.

Requirements

You can browse demo content without signing in, but you must be signed in to generate images.

Local GPU generation requires Pro access, a running local server, and at least one compatible local model installed. Local GPU generation is shown as Pro in the estimate area.

Typical workflow

  1. Open AI Image Gen.
  2. Choose Cloud or Local GPU in the left panel.
  3. Select one or more models.
  4. Enter a prompt in the right panel.
  5. Choose an aspect ratio.
  6. Choose a quantity from 1 to 4.
  7. Add reference images if the selected model supports them.
  8. Review the estimate area.
  9. Click Generate.
  10. New generation cards appear at the top of the gallery while they process.

If Local GPU Save generated images is enabled, finished local images are uploaded to your library. If it is disabled, local results remain visible only in the current page session and disappear when the page refreshes.

Cloud mode

Cloud mode is the default workspace. It supports selecting multiple models at the same time, which is useful when you want to compare how different models interpret the same prompt.

Selecting cloud models

Use the model picker in the left panel to select cloud models. On mobile, tap Models to open the picker. The selected model count appears in the mobile toolbar.

Some cloud models include a settings control. The exact settings vary by model, but the UI may show:

  • Dropdown options.
  • On/off toggles.
  • Numeric sliders.
  • Short descriptions for available settings.

Model settings are saved with the generation and restored when you reuse that image.

Aspect ratio compatibility

Aspect ratio buttons are filtered by your selected cloud models. A ratio is available only when all selected cloud models support it. Unsupported ratios appear disabled. If your model selection conflicts with the current ratio, Image Gen may show a warning and ask you to choose a compatible ratio.

Local GPU mode does not use the same cloud aspect-ratio compatibility filtering, so the aspect buttons remain available while local mode is selected.

Cloud generation estimate

The estimate at the bottom of the prompt panel updates based on:

  • Selected model or models.
  • Quantity.
  • Aspect ratio.
  • Model-specific settings.
  • Whether reference images are being used.
  • Models that return more than one output for a single request.

Local GPU mode

Local GPU mode lets Pro users generate with supported local models through the Sorceress local server.

Image Gen always opens in Cloud mode. If you used Local GPU previously, switch back to Local GPU manually from the left panel.

Local Server card

In Local GPU mode, the left panel shows a Local Server card. It automatically checks for your local server when the page loads and displays whether the server is connected.

When a server is detected, the card can show the currently running model/profile. If the running local profile can be matched, clicking the running status selects it in Image Gen.

Available controls include:

  • Refresh — re-checks the local server and installed model information.
  • Start — opens the local manager when the server is not running.
  • Install — opens model installation options when the server is running.
  • Config — opens server settings and install options when available.
  • Console / Debug Logs — opens live server output for troubleshooting.
  • Unload Model · Free RAM — unloads the currently loaded model from memory when no local generation is running. Installed model files remain on disk.

If the server is offline, local models are shown but locked. Start the server, then click Refresh.

Selecting local models and variants

Local models may have multiple installable variants. When a model has variants, use the gear control next to it to choose among them.

Variant entries can show:

  • Which variant is currently selected.
  • Which variant is currently running.
  • Whether a variant is installed.
  • Hardware or storage guidance such as VRAM or disk requirements.

If a model family or variant is not installed, clicking it opens the local manager so you can set it up. Variants marked Not installed also open the manager.

Local GPU can select more than one installed local model family. The Generate button shows the total number of images based on selected local models and quantity.

Local reference behavior

Local models differ in how they handle references:

  • Some are text-only and hide or reject reference image input.
  • Some support optional reference images.
  • Some are image editors and require a reference image.
  • Some text-to-image models automatically switch to a related editing model when you add a reference image.

The local model list labels models as Text only, Optional reference, Reference required, or Editing with reference depending on the selected model and whether a reference has been added.

If a selected local editor requires a reference, the Generate button changes to Add a reference image to edit until you add one. If you try to use references with a text-only local model, Image Gen shows a warning and may remove unsupported references.

When a local model has an automatic reference-editing companion, the reference area explains whether adding a reference will switch the model for editing.

Saving local generations

Local GPU mode includes a Save generated images toggle near the Generate button:

  • On: finished local images are uploaded to your library and remain available later.
  • Off: generated local images display in the current page session only. They are not saved and disappear when the page refreshes.

This is useful for quick experiments when you do not want to keep every local result.

Prompt controls

The prompt field accepts the description of the image you want to create. The text area expands for longer prompts.

If your browser supports speech recognition, a microphone button appears near the prompt label:

  1. Click the microphone to start voice input.
  2. Speak your prompt.
  3. Click again, or start generation, to stop listening.

While voice input is active, Image Gen shows a Listening... speak now indicator.

Compatible Sorceress links may open Image Gen with a prompt, model, aspect ratio, or resolution already filled in.

AI prompt agent

When using Cloud mode, the left panel includes an AI prompt agent panel. Use it to help create or refine a prompt, then send the generated prompt directly into Image Gen. When the agent triggers generation, the prompt field updates so you can see what was used.

The agent panel is not shown in Local GPU mode.

Reference images

When the current model selection supports references, a Reference Images area appears in the prompt panel. If no selected model can use references, this area is hidden.

You can add references by:

  • Clicking the upload area and choosing image files.
  • Dragging image files into the upload area.
  • Dragging completed images from the gallery.
  • Dragging compatible Sorceress image or tile items into the page.
  • Dropping image files from WizardGenie Explorer when used in a compatible embedded context.

Reference images are resized before generation. The UI shows how many reference images are currently added and the maximum supported by the selected model selection. If you add too many, Image Gen warns you and visually marks references beyond the limit.

Reference thumbnails include controls to:

  • Open the reference in a full-size lightbox.
  • Download the reference.
  • Remove the reference.

The reference lightbox supports:

  • Previous/next navigation.
  • Keyboard arrow navigation.
  • Escape to close.
  • Download.
  • Remove.

The center panel shows your Image Gen gallery. New generations appear at the top. Cards may show a spinner while processing or an error panel if generation fails.

Gallery tools include:

  • Search prompts — filters visible images by prompt text.
  • Grid — adjusts desktop gallery column count from 1 to 5.
  • All / Favorites — switches between all images and favorited images.
  • Select — enters multi-select mode for bulk actions.
  • Infinite scrolling for older generations.

If you are not signed in, the page may show demo images. Generation still requires sign-in.

Working with a generated image

Click a completed image to open it in the lightbox. From the lightbox you can:

  • View the full image.
  • Navigate to previous or next completed image using buttons or keyboard arrows.
  • See the model and aspect ratio.
  • Read the prompt.
  • Reuse the prompt and settings with Use Prompt.
  • Copy the prompt.
  • Download the image.
  • Send the image to another tool.
  • Drag the image to WizardGenie Explorer when embedded in WizardGenie.
  • Close with Escape or the close button.

Hovering a gallery card reveals quick actions:

  • Favorite — mark or unmark the image.
  • Share to Prompt Lexicon — share the prompt, image, optional references, and an optional comment.
  • Download — save the image to your device.
  • Reuse — load the prompt, aspect ratio, model, settings, and references back into the generator when available.
  • Send to... — pass the image into another Sorceress tool.
  • Delete — remove the image from your library.

Failed cards show the error message and a Remove button.

Generating cloud cards may show a Recheck action after they have been running for a while. Use it if a generation appears stuck. Local generation cards can show local progress and status text.

Reusing a generation

Click Reuse on a gallery card or Use Prompt in the lightbox. Image Gen restores:

  • The prompt.
  • The aspect ratio.
  • The model selection.
  • Model-specific settings.
  • Reference images, when the original references are still available.

The page scrolls to the top so you can edit and generate again.

Sending images to other tools

Use Send to... from a gallery card or from the lightbox to move a completed image into another Sorceress workflow.

Available destinations include:

  • VideoGen
  • BG Remover
  • Expander
  • Canvas
  • Slicer
  • True Pixel
  • Tileset Forge
  • Sprite Analyzer
  • AutoSprite

You can also use multi-select mode to send multiple completed images to AutoSprite.

When AI Image Gen is embedded in WizardGenie, completed images can be dragged into WizardGenie Explorer.

Favorites and collections

Click the star button on a completed image to favorite it. Use Favorites in the gallery toolbar to view favorited images.

When viewing favorites, the Collections manager appears below the gallery. It lets you browse image collections and open a collection view. Use Back to Favorites to return from a collection to the favorites list.

Multi-select, bulk download, and bulk delete

Click Select in the gallery toolbar to enter batch selection mode. The action bar appears directly below the gallery toolbar.

In batch mode you can:

  • Select individual completed images.
  • Select All visible completed images.
  • Clear the current selection.
  • Send to AutoSprite.
  • Choose a download format.
  • Download selected images.
  • Delete selected images.
  • Click Done to exit select mode.

Bulk download format options are:

  • Original
  • PNG
  • JPG
  • WebP

Bulk delete asks for confirmation and cannot be undone.

Sharing to Prompt Lexicon

Use the share button on a generated image to open the Share to Prompt Lexicon dialog.

The share dialog includes:

  • The generated image preview.
  • The prompt.
  • The image model name.
  • A reference-image badge when references were used.
  • An optional comment field up to 500 characters.

Click Share to Lexicon to publish, or Cancel to close without sharing. After a successful share, the share button changes to show that the image has been shared during the current session.

Troubleshooting

“Sign in to generate” appears

You are not signed in. Click the button and sign in before generating.

Your selected cloud generation cannot be started with the current account state. Reduce quantity, select fewer models, change options, or use the dialog to get more.

Local GPU says the server is offline

Click Start to open the local manager, start the server, then click Refresh. Models remain locked until the server is reachable.

A local model is not selectable

The local server may be offline, the install check may still be running, or the model/variant may not be installed. Start the server, refresh, then install the model or choose an installed variant.

Generate is disabled in Local GPU mode

Common causes:

  • No local model is selected.
  • A selected editor model requires a reference image.
  • You are not signed in.
  • Your account does not have Pro access.

Reference images disappear or are rejected

The selected model may not support references, or you may have exceeded the selected model’s reference limit. Choose a model that supports references or remove extra images.

A cloud generation is taking a long time

Wait a little longer, then use Recheck if it appears. If the generation cannot be recovered, it may be marked failed.

A local generation seems stuck

Local GPU generations can take several minutes, especially when loading a model for the first time. Open Console / Debug Logs to watch server output. If needed, wait for the job to finish, then use Unload Model · Free RAM after generation completes.

A local generation fails with a long error message

Open Console / Debug Logs in Local GPU mode. Failed local cards may include detailed server log output, which can help identify missing models, server problems, or hardware limitations.

I want to free memory after using Local GPU

If a model is loaded and no generation is running, click Unload Model · Free RAM in the Local Server card. This frees RAM/VRAM without uninstalling the model.