Read the interface from model to task to prompt to settings to output. That order separates a missing model, the wrong workflow, an unsuitable generation value, and a result-handling problem before they blur into one “Forge is not working” report.
The first screen feels dense because model loading, generation, optional tools, and result routing share one browser application.
Confirm UI, Checkpoint, and only the required VAE / Text Encoder files.
Use Txt2img for text-to-image, Img2img for a source image, or the specialized tab that owns the job.
Enter Prompt and, only when the model workflow uses it, Negative prompt.
Copy the exact model’s sampler, scheduler, steps, size, guidance, and component requirements.
Keep optional accordions out of a clean baseline and watch the console as well as browser progress.
Check the gallery, infotext, saved file, effective folder, and console result.
Exact labels and construction order are traced to Original Forge source. Placement can change with screen width, UI options, extensions, and commit.
The original Forge runtime was not launched for this review, and the clean repository does not contain a current product screenshot. This artifact maps verified code labels and relationships without claiming pixel-exact placement.
Read UI preset, Checkpoint, and VAE / Text Encoder before touching generation controls.
Txt2img starts from text. Img2img starts from an image. Extras post-processes an existing result.
Prompt and Negative prompt sit beside Generate, with Interrupt and Skip used after work begins.
Sampler, scheduler, steps, dimensions, guidance, batch, seed, and Hires. fix describe the run.
Scripts and bundled or third-party extensions can insert accordions and controls into Txt2img or Img2img.
The gallery, infotext, folder, Save, ZIP, and send-to actions serve different purposes.
VERIFIED: labels, choices, callbacks, and interface construction at dfdcbab. UNKNOWN: pixel layout in an unlaunched local runtime. VARIABLE: theme, viewport, saved UI settings, extensions, and later commits.
These controls define the model side of the run. They are upstream of prompts, sampling settings, and optional guidance panels.
sd · xl · flux · allShows a family-oriented control surface and starting values. Choose all when you need every top memory control visible. It is not a model selector.
The list proves Forge discovered a supported file. Only a successful load and generation prove that the complete file set works.
Fix discovery →This is a multiselect. Use separate components only when the model publisher requires them; a shared .safetensors suffix does not make files interchangeable.
The adjacent refresh action updates Checkpoint and VAE / Text Encoder choices. It cannot repair a wrong path, corrupt file, or incompatible family.
Run the file check →These change precision or weight placement, not image composition. Leave a working baseline intact and use the dedicated memory procedure when memory pressure gives you a reason to change them.
Open the memory controls guide →Changing the tab changes the input contract. It does not merely reveal another group of preferences.
Generate from a text prompt.
Transform an input image; inpainting modes live inside this task.
Launch separately managed mini-app wrappers.
Post-process an existing image, including upscaling.
Read embedded generation information and route recovered settings.
Combine compatible checkpoints; provenance and licenses remain separate.
Change persistent WebUI and extension options.
Install, update, disable, or remove extension code.
VERIFIED core construction at dfdcbab. Extension callbacks run before Settings and Extensions are appended; saved UI tab order and Hidden UI tabs options are applied when the outer tab set renders.
The prompt box describes the request. The action box controls the current generation job.
Describe what the active model should generate. Plain language is the cleanest first diagnostic.
Use it only when the model’s guidance path supports it. At CFG Scale 1, inspected Txt2img code disables this field.
Submit one job. Right-click behavior may expose an additional continuous-generation menu.
Stops the current batch and continues to the next batch in the job.
Ends generation now or after the current image when the corresponding setting and job state apply.
Use the hover tooltip to distinguish Read generation parameters, Clear prompt, Apply selected styles, and Restore progress. Img2img adds CLIP and DeepBooru interrogation actions. Icon appearance can change; the tooltip is the safer label.
Inspect the action code ↗Forge presents a common control vocabulary across model families. The exact model documentation remains the authority for a starting recipe.
Which numerical sampling method advances the denoising process.
Read the canonical guide →Schedule typeHow noise levels or timesteps are distributed across the requested steps.
Read the canonical guide →Sampling stepsHow many denoising steps the run requests; more is not automatically better.
Read the canonical guide →Width · HeightOutput geometry. Begin with the exact model publisher’s supported baseline.
Read the canonical guide →CFG · Distilled CFGSeparate guidance paths; which one matters depends on the model.
Read the canonical guide →Batch count · Batch sizeRepeated batches versus images processed together. Batch size usually pressures peak memory more directly.
Read the canonical guide →SeedThe starting-noise number. Use the concrete result seed when comparing one controlled change.
Read the canonical guide →Hires. fixAn optional second diffusion pass with its own additional settings, not a first-run requirement.
Read the canonical guide →Forge lets active scripts attach controls to named UI sections. Bundled and user-installed extension state therefore changes what appears below the core settings.
The inspected constructor creates this accordion directly, with second-pass upscaler, denoising, resolution, guidance, and optional model controls.
Choose a two-pass or post-process route →Active scripts can insert always-visible accordions into Txt2img and Img2img. Their presence and position depend on the loaded script set.
Use ControlNet deliberately →Installed extensions can add accordions, tabs, settings, JavaScript, and backend code. A copied screenshot does not establish compatibility.
Test extension ownership →Result actions appear together, but they write, reveal, copy, or route different artifacts.
Select the final image you want to inspect or route. A preview appearing does not by itself prove the configured file was saved.
Asks Forge to open the effective sample directory, subject to launch restrictions and output settings.
Copies selected result files to the dedicated Save location or builds a ZIP. This is separate from automatic sample saving.
Shows generation parameters for the selected result. Keep model and extension files separately; metadata cannot embed the whole environment.
Sends the selected image and registered parameters to another task without making missing dependencies follow it.
Txt2img can expose an action that uses Hires. fix settings to create an upscaled version of the current image.
These are administration surfaces for the current installation, not merely more generation controls.
Stores applicable option changes. Some options run a callback; others explicitly need the interface to reload before presentation changes.
Reconstructs the interface. Use it after changing compact prompt layout, tab order, hidden tabs, and other options marked as requiring reload.
Configured settings can be rendered in a quick-settings row. Their presence depends on the saved quick-setting list and active tab.
Install, update, disable, and remove extension code here. Treat every new extension as a change to both UI and runtime state.
Install and verify one extension →Back up the active configuration and record the commit before broad changes. “Reload UI” is not a rollback, and an extension can fail during reconstruction even if installation completed. Use the extension operations guide before adding third-party code.
Start with the smallest state difference that can explain the missing or moved control.
Select the all UI preset to expose the complete top control set at the inspected commit. Do not change GPU Weights merely because it became visible.
Check Settings for UI tab order and Hidden UI tabs, then reload the UI. If the tab is not one of the eight code-constructed core tabs, identify the extension callback that supplies it.
The Txt2img settings accordion and Img2img settings accordion options can wrap the relevant controls under Open for Settings. Expand it or review that saved presentation setting.
Compact prompt layout moves prompts into the generation area and renders the action box beside the output. Screen width and theme can also reflow the layout without changing the job.
Use the model-file location procedure. Verify the effective data/model directory and launch overrides, refresh, then restart once. A prompt or sampler change cannot make an undiscovered file appear.
Resolve file discovery →Record the extension repository, version or commit, Forge commit, and first console error. Disable the suspected extension for a clean baseline instead of attributing its UI to Original Forge.
Isolate the extension →Return to the console and the earliest failed stage. Separate model load, sampling, VAE decode, browser rendering, and file saving before changing controls.
Locate a black or blank output failure →These answers cover navigation and ownership. Dedicated model, workflow, performance, and troubleshooting pages handle the deeper task.
Generate belongs to the prompt action area in Txt2img and Img2img. A compact prompt-layout setting can move the prompt and action box within the page, so look in the active generation tab rather than comparing pixel positions with a tutorial.
First confirm the UI preset, Checkpoint, and any required VAE or Text Encoder. Then open the task tab. For a first text-to-image result, use Txt2img, enter a plain prompt, copy the model’s documented baseline, and select Generate once.
UI offers sd, xl, flux, and all in the inspected source. A preset changes which model-oriented controls are visible and supplies family-oriented starting values. It does not download a model, convert one family into another, or prove file compatibility.
GPU Weights is part of the advanced global memory controls, and its visibility depends on the UI preset. The inspected code itself advises selecting all when that control is not visible. Read the memory guide before changing the value.
Checkpoint is in the global model bar near UI and VAE / Text Encoder. If its list is empty, verify the effective model directory and launch overrides, use the adjacent refresh control, and solve discovery before editing generation settings.
It is a multiselect for additional model components. Some layouts bundle these components; split layouts can require separate VAE, CLIP, or T5 files. Select only the files required by the exact model documentation.
They belong to the Generation area of Txt2img or Img2img. UI reordering, a compact Settings accordion, and script-provided sections can change the vertical order, but these controls describe the run rather than the loaded model.
In the inspected Txt2img code, changing CFG Scale to 1 makes Negative prompt non-interactive. That is a code behavior, not a recommendation to use CFG 1 for every model. Follow the model’s guidance contract.
When the relevant preference is enabled, Forge wraps generation controls in an accordion labeled Open for Settings. Expand it to reach those controls; the setting changes presentation, not the underlying generation task.
Txt2img begins from text conditioning and a generated latent. Img2img takes a source image and uses denoising strength to control how far the result can move from it. Inpaint, sketch, upload, and batch modes are nested within Img2img in the inspected interface.
Settings exposes UI tab order and Hidden UI tabs, both requiring a UI reload. Extensions can also register their own tabs. Compare repository, commit, settings, and enabled extensions before treating a screenshot difference as a bug.
First identify whether the panel is built in, bundled as an extension, or supplied by a third party. Its visibility can depend on script loading, extension state, current task, and commit. Do not assume a panel shown in an old video is part of every Original Forge build.
The inspected tool row includes actions to read generation parameters, clear the prompts, apply selected styles, and restore progress when that hidden recovery action is available. Img2img also exposes CLIP and DeepBooru interrogation tools. Read the tooltip before selecting an icon.
Skip stops the current batch and continues to the next queued batch. Interrupt ends generation immediately or, depending on the Don’t Interrupt in the middle setting and job count, after the current image. If you generated only one image, there may be no later batch to skip to.
The Output gallery shows browser results. Forge can also automatically save samples to its configured output directory. The folder button opens the effective output location, while Save copies the selected result to a separate dedicated save directory.
The output action row registers send-to buttons for Img2img, Inpaint, and Extras. These route the selected gallery image and relevant generation parameters; they do not copy missing model or extension files.
Settings contains separate Apply settings and Reload UI buttons. Apply stores applicable changes; options such as tab order, hidden tabs, and some layout settings explicitly require a UI reload. Keep the console visible in case reload exposes an extension error.
The map is bound to Original Forge commit dfdcbab and is derived from source labels, not a runtime screenshot. A different commit, compact-prompt setting, hidden or reordered tabs, model preset, screen width, theme, or extension set can change the screen.
Repository code establishes labels and behavior. Dated videos establish user vocabulary and confusion, not current product truth.
Labels, choices, callbacks, tab construction, visibility options, output actions, and reload controls were checked in the original repository.
Four complete video transcripts supplied navigation questions. Their performance, compatibility, and feature-state claims were excluded.
Forge was not launched with models in this environment. No generated mockup is presented as a product screenshot.
Exact global labels, choices, refresh behavior, and preset-controlled visibility.
V01 · STALE SNAPSHOTFull install and run videoUsed only to collect beginner wording and navigation questions from its transcript.
V04 · STALE SNAPSHOTLong-form Forge walkthroughUsed only to identify where users hesitate around model controls, generation settings, and extensions.
V08 · STALE SNAPSHOTInstall and configuration walkthroughUsed only to collect user language for checkpoints, prompts, settings, and output.
V11 · STALE SNAPSHOTBeginner UI walkthroughUsed only to identify tab-discovery intent; dated feature and performance claims were excluded.
Freshly rechecked top-level tabs, generation sections, Img2img modes, and extension tab callbacks.
OPEN SOURCE ↗Current prompt action rowFreshly rechecked Prompt, Negative prompt, Generate, Interrupt, Skip, shortcuts, and tooltips.
OPEN SOURCE ↗Current Forge model barFreshly rechecked UI preset, Checkpoint, VAE / Text Encoder, refresh, and memory-control labels.
OPEN SOURCE ↗Commit-bound output panelVerified gallery, folder, Save, ZIP, infotext, and send-to actions.
OPEN SOURCE ↗Commit-bound settings UIVerified Apply settings, Reload UI, and quick-settings construction.
OPEN SOURCE ↗Commit-bound UI optionsVerified compact prompt layout, UI tab order, hidden tabs, and reload requirements.
OPEN SOURCE ↗Written by the Forge Field Guide editorial team · technically reviewed against Original Forge commit dfdcbab · evidence checked 1 Sep 2026 · runtime UI not executed.