Multi-modal image editing
Systems and methods for multi-modal image editing are provided. In one aspect, a system and method for multi-modal image editing includes identifying an image, a prompt identifying an element to be added to the image, and a mask indicating a first region of the image for depicting the element. The system then generates a partially noisy image map that includes noise in the first region and image features from the image in a second region outside the first region. A diffusion model generates a composite image map based on the partially noisy image map and the prompt. In some cases, the composite image map includes the target element in the first region that corresponds to the mask.
1. A method for image editing, comprising:
identifying, by a processor, an image, a prompt identifying an element to be added to the image, a preliminary mask, and a mask precision indicator;
expanding, by the processor, the preliminary mask based on the mask precision indicator to obtain a mask indicating a first region of the image for depicting the element;
generating, by the processor, a partially noisy image map that includes noise in the first region and image features from the image in a second region outside the first region; and
generating, by the processor, a composite image map using a diffusion model based on the partially noisy image map and the prompt, wherein the composite image map includes the element in the first region that corresponds to the mask.
2. The method of claim 1 , further comprising:
receiving a brush tool input from a user; and
generating the preliminary mask based on the brush tool input.
3. The method of claim 1 , further comprising:
encoding the image to obtain an image feature map, wherein the partially noisy image map is based on the image feature map; and
decoding the composite image map to obtain a composite image.
4. The method of claim 1 , further comprising:
generating a predicted mask for the element based on the composite image map; and
combining the image and the composite image map based on the predicted mask.
5. The method of claim 1 , further comprising:
providing the mask as an input to the diffusion model.
6. The method of claim 1 , further comprising:
generating intermediate denoising data using the diffusion model; and
combining the intermediate denoising data with the partially noisy image map to obtain an intermediate composite image map.
7. A system comprising:
a memory component; and
a processing device coupled to the memory component, the processing device configured to perform operations comprising:
identifying an image, a prompt identifying an element to be added to the image, a preliminary mask, and a mask precision indicator;
expanding the preliminary mask based on the mask precision indicator to obtain a mask indicating a first region of the image for depicting the element;
generating a partially noisy image map that includes noise in the first region and image features from the image in a second region outside the first region; and
generating a composite image map using a diffusion model based on the partially noisy image map and the prompt, wherein the composite image map includes the element in the first region that corresponds to the mask.
8. The system of claim 7 , further comprising:
a user interface that includes a brush tool for generating the mask and a precision control element for obtaining a mask precision indicator.
9. The system of claim 7 , wherein:
the diffusion model is further configured to generate image features based on the image, and decode the composite image map to obtain a composite image.
10. The system of claim 7 , wherein:
the diffusion model comprises a U-Net architecture.
11. The system of claim 7 , wherein:
the diffusion model comprises a channel corresponding to the mask.
12. The system of claim 7 , the processing device being further configured to perform operations comprising:
generating a predicted mask based on the composite image map using a mask network.
13. A non-transitory computer readable medium storing code for data processing, the code comprising instructions that, when executed by at least one processor, cause the at least one processor to perform operations comprising:
identifying an image, a prompt identifying an element to be added to the image, a preliminary mask, and a mask precision indicator;
expanding the preliminary mask based on the mask precision indicator to obtain a mask indicating a first region of the image for depicting the element;
generating a partially noisy image map that includes noise in the first region and image features from the image in a second region outside the first region; and
generating a composite image map using a diffusion model based on the partially noisy image map and the prompt, wherein the composite image map includes the element in the first region that corresponds to the mask.
14. The non-transitory computer readable medium of claim 13 , the code further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
receiving a brush tool input from a user; and
generating the preliminary mask based on the brush tool input.
15. The non-transitory computer readable medium of claim 13 , the code further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
encoding the image to obtain an image feature map, wherein the partially noisy image map is based on the image feature map; and
decoding the composite image map to obtain a composite image.
16. The non-transitory computer readable medium of claim 13 , the code further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
generating a predicted mask for the element based on the composite image map; and
combining the image and the composite image map based on the predicted mask.
17. The non-transitory computer readable medium of claim 13 , the code further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
providing the mask as an input to the diffusion model.
18. The non-transitory computer readable medium of claim 13 , the code further comprising instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
generating intermediate denoising data using the diffusion model; and
combining the intermediate denoising data with the partially noisy image map to obtain an intermediate composite image map.