Overview

Reforge allows users to enhance their image generation process by incorporating input images. Users can upload images to serve as references during the prompting phase. These reference images can influence the style of the generated images, ensuring they align closely with desired visual aesthetics.

Additionally, users can integrate Face IDs from uploaded images into the image generation process. This feature enables the AI to incorporate specific facial characteristics or non-facial attributes from the uploaded images into the generated outputs.

Moreover, the Reforge feature offers versatile functionalities for image manipulation:

i) Style Transfer: Users can replicate the visual style of uploaded images, ensuring the generated content matches the desired artistic or thematic qualities.
ii) Pose Mimicking: It allows for mimicking the poses of individuals depicted in uploaded images, ensuring the generated figures adopt similar stances or gestures.
iii) Replication: Users can replicate entire images tile by tile, while integrating the unique facial features retained by a fine-tune.

Using Face IDs

Using Face IDs allows you to do a virtual try-on of a garment that you like on a model that you have fine-tuned to look like you. This feature is particularly useful for fashion designers, clothing brands, and individuals who want to visualize how a particular outfit would look on them before making a purchase. FaceIDs don't train the model and as such are ready immediately for use.

Mask Prompting

Mask prompting allows users to specify regions of interest within an image, guiding the AI to focus on particular elements during the generation process. By masking specific areas, users can control the output's details, style, and composition, ensuring the final image aligns with their creative vision.

Insert comma separated words here that you want to mask out of the image. This is especially useful when looking to redecorate a room and you want to retain the location of doors and windows in the room.

Mask Invert

Enabling Mask Invert will invert the mask selection of the words inputted as part of Mask Prompting.

Control Net (Input Image Required)

Control net enables the preservation of composition of images inputted by the user. Following are the modes made available through control net as of Photocobra 1.0.0:

i) Style - Will preserve the art style of the input image. This can be used when trying to recreate a painting in the style of a famous artist.
ii) Pose - Will preserve the pose of the subject in the input image. This can be used when trying to recreate a pose of a person in a different setting.
iii) Tile - Will preserve all the details of the input image except the details tuned on the fine-tune (for e.g a face will not be preserved from the image but will come from the fine-tune instead). This can be used when trying to recreate a scene in the style of the input image.

Enable Text to Image if you’d like to create more variation and draw more from the prompt instead.

Logo Reconstructions
Control Net can be used to create logo reconstructions where you can pass in the original logo as an input image.
i) Insert the original logo as the input image. Black and White images are recommended when passing in the original logo.
ii) Choose the control net as QR and enable Text to Image to draw more from your prompt
iii) Insert a prompt describing the setting that you want the logo to appear in.
iv) Use additional features from the advanced section to further enhance the image.

Virtual Try-on:
i) Fine-tune a model (Realistic Vision v5.1 recommended) on a dataset of your images. Ensure that the dataset is appropriately varied and representative of your appearance.
ii) Generate a face-id fine-tune of the garmet/outfit you want to try on. And train it under the class of shirt, pants, clothing, etc. depending on what aspect of the image you're trying to capture. This image can be that of a model wearing the outfit, or a plain image of the outfit laid flat on a surface, or even an image of the outfit donned on a mannequin.
iii) Head over to reforge, and choose the face-id of the garment you want to try on, and select the model that will be wearing this outfit. Add an additional prompt (See Prompting) for some extra context, and generate the image.