Choose a look before you describe it
A clothing reference is useful when you can recognize the outfit you want but do not know its name. Instead of listing every sleeve, layer and fastening, choose an image that shows the complete look. COSTUME-DB-01 provides visual clothing references you can browse before bringing a single outfit into this tool.
Start with the silhouette: a fitted dress, an oversized jacket, a layered skirt or a matching set. Then compare the color, fabric and accessories. Pick the reference closest to the result you want. The generator uses the image itself, so a different design with a similar description will still lead to a different outfit.
One image supplies your character
Your character image supplies the person, face, hairstyle and visual style. A clear full-body illustration makes a useful starting point for a complete outfit. A portrait can work for a top or neckline, but it gives the model less information about the character's lower body and existing composition.
Use the character you want to keep. This may be an original character, an existing AI model or your own photo. The default instructions keep the character's gender and body proportions as well as their face. Choosing a dress does not automatically request a gender transformation. You can edit those instructions when your creative goal is different.
The other image supplies the outfit
Add one clothing reference in the outfit field. A garment-only image makes the division clear: the character comes from the first image, the clothes from the second. If your reference includes another model, the default instructions ask for that person's clothing without copying their identity, hairstyle or body.
Keep the parts you care about visible. For a layered look, the jacket, inner top and lower layer should be distinguishable. For a complete outfit, include the hem and shoes. A screenshot of a gallery grid contains several competing outfits; open and save the single item before uploading it here.
Use the instructions to set priorities
The tool starts with editable instructions that assign a role to each image. Keep my character preserves the original pose and setting. Match outfit details asks the model to pay attention to the neckline, sleeves, seams and accessories. Full-body look asks for the entire character, which can also change the framing when your source is a close portrait.
Choose a suggestion as a starting point, then add the detail that matters to your project. For example, keep the character's glasses, leave the background unchanged, or use the dress without the bag. Suggestions fill the instructions only; they do not submit the images or spend credits. Check the text before generating.
Try another outfit with the same character
For a series of looks, leave the character image in place and replace only the outfit reference. You can compare a casual look, a seasonal outfit and a special-occasion costume against the same starting character. Keeping that source image gives every attempt the same identity reference instead of accumulating changes from earlier outputs.
Continue editing has a different purpose: it replaces the character input with your generated result so you can refine that version. Use it when you like the new outfit and want to adjust a detail. Simply selecting another garment keeps your earlier result visible until a new generation succeeds.
Check identity and clothing separately
Compare the output with both references. First check the character's eyes, hair, facial structure and any distinctive features. Then check the clothing's overall shape, color, sleeve length, neckline and layers. Small buckles, lace patterns, jewelry and logos can be simplified or changed even when the overall look is recognizable.
When one part is wrong, make the next request specific. Ask to preserve the current face while correcting the sleeve, or return to your original character with a clearer outfit reference. The output is a creative image preview. It does not establish garment measurements, shopping size or how a physical piece will fit.