Hi, how can we help you today?
Hi, how can we help you today?
The Image Generation interface (link) is the primary tool for creating images and content in Scenario. It’s built to generate consistent visuals through a structured approach using models, prompts, or reference images.
You can generate inferences (the process of creating images) by selecting models, prompts, reference images, and settings, adjusting parameters in the right column. Once generated, the results appear on the left, where you can refine their display and iterate further. Here’s how it works.
Access Image Generation - Choose one of three ways:
In the Scenario web app, open the left panel, select “Images“, and click “New”.
From the Image gallery, click “Generate Images“ (top right). For both option 1 and 2, you’ll have to pick a model to proceed.
On a model page, hit “Use this model” > “Generate” to load the interface with that model ready.
The inference panel is where you configure and adjust all the parameters for your generation before starting the process. Here’s a breakdown of its elements, from top to bottom:
1 - Model
Every inference requires a model, either a “base model” (or “foundation model”) like Flux, or a custom-trained model, such as a LoRA or multi-LoRA tailored to specific styles or subjects.
Opening the model menu brings up a modal with a list of available models, along with filters to help you browse or search for the right one. For more details on selecting or creating models, refer to the Models section in the Knowledge Base.
When using a multi-LoRA, you can adjust the influence of each individual LoRA component under the Model Components section. If using a single-LoRA, only that specific LoRA will be displayed.
2 - Prompt
The prompt is your description of the content you want to generate and serves as a key tool for guiding the model. Depending on the model, prompts can be either brief or highly detailed. Since different models will interpret the same prompt in varying ways, it's important to tailor your input per the model, for the best results.
Scenario has developed a set of tools to help you craft effective prompts, called “Prompt Spark“. Use it to generate a solid prompt from scratch (tailored to the model used), expand on a few initial words/ideas, or even translate a prompt from any language before refining it. Learn more here.
Stable Diffusion models also support Negative Prompts, letting you specify what to exclude, like objects, styles, or colors.
3 - Reference Images
Beyond prompts, you can guide the AI further by uploading reference images. Different modes extract specific features from these reference images (such as the colors, depth, or style for instance). You can use one or two reference images, each with its own mode. Reference images are a very powerful tool to better guide your generations. Explore the available modes here: [3.3].
You can upload an image from your computer, choose one from your library, or sketch directly within Scenario. Alternatively, you can simply copy and paste an image from your clipboard, and it will automatically load into the menu. Once added, use the slider beneath the image to adjust its influence. Different modes may require varying influence levels to achieve the right balance between consistency and creativity.
Keep in mind that image influence is part of a set of parameters: its effect depends on the model, the prompt, and the selected settings. Adjust accordingly to find the right balance.
4 - Settings
Additional inference settings and options include:
Image Count: Select how many images to generate per inference.
Sampling Steps (also known as denoising steps) refer to the number of “passes” the AI model takes to refine and generate the image. More steps generally result in higher detail and quality, but also increase inference time. The default value provides a balanced trade-off between speed and quality.
Guidance: Controls how closely the image follows your prompt. Adjust the scale to tweak its influence.
Dimensions: Set the aspect ratio for your generated images.
Scheduler: A technical setting that subtly affects the generation process alongside the model.
Seed: A key setting that controls the randomness in image generation. It acts as a unique identifier for the generated result, ensuring that the same prompt, model, and settings will consistently produce identical images when using the same seed. Changing the seed introduces variation, allowing for different outputs while keeping other parameters unchanged.
5 - Generate
When all your parameters are set, simply hit the blue “Generate“ button to start the inference and generate your content. The total cost (in credits) is displayed on the blue button. Use Cmd/Ctrl + Enter as a faster shortcut. Press “H” to see a full list of shortcuts available in Scenario.
The content panel is where your generated images appear once an inference is complete. Click on any image to enlarge it, view its generation settings, or make further edits and refinements.
Key Features of the Content Panel:
Filters (1) in the top left let you customize the displayed elements, allowing you to focus on specific authors or actions, such as generated images, enhancements/upscales, background removal, and more.
Clear Session (2): This button lets you remove images from the current session without however deleting them from your library (to delete, select instead one or more images and click “Delete“).
Placeholders (3): Before generating, grey placeholders show the size and number of upcoming images.
Zoom Level (4): Slide to resize how many thumbnails are previewed on-screen. Or use the +/- keys as shortcuts.
“Display Alignment“ options (5): Choose from four modes to arrange images for easier viewing and comparison:
Masonry: Images are staggered like bricks in a wall.
Fill in Grid: Images are zoomed in to fully occupy a uniform square grid.
Fit in Grid: Images resize proportionally to fit within a square grid without cropping.
Row (by Inference): Images align in rows based on each generation.
The generation interface is built for step-by-step refinement, reducing the unpredictability of relying solely on prompts. You can configure Inference “A” on the left, generate images and inspect them on the right, then adjust parameters to create Inference “B”, allowing you to compare results side by side, and iterate as needed.
To reuse elements from a previous inference, such as the prompt, model, or seed, click the three dots in the top right of any generated image. From the menu, you can select:
All Parameters: Reuse the full set of settings from the selected image.
Model: Apply the same model used in the inference.
Image: Use the image as a reference for a new generation.
Prompt: Copy and reuse the original prompt.
Seed: Maintain the same seed for consistency in results.
Additional options allow you to edit and refine images:
Enhance: Improve resolution & image quality.
Retouch: Make manual, detailed adjustments (inpainting).
Expand: Extend the image beyond its original boundaries (outpainting).
More: Access additional editing tools.
For exploring variations, the Search menu lets you find:
Similar Images: Search for images with a similar style or composition.
To reuse an image only, you may either Select “Reuse Image” to load it as a reference image, in the left inference panel. Or Drag and drop the image from the content panel into the reference image menu, either as a single reference or alongside another image for dual-reference guidance.
You can utilize Scenario's API to generate images from text prompts by following these steps:
Run a Text-to-Image (txt2img) Inference: Submit a text prompt to create an image.
Retrieve Inference Results: Obtain the generated images from your inference request.
Resources:
Run a txt2img Inference and Retrieve Results: Scenario API Recipe
Text-to-Image Inference Endpoint: Scenario API Documentation - POST /txt2img/inferences
Was this helpful?
Quentin