[Scribble] How to catch the color in input image?

[DevJunghun]

Hi, Thanks for sharing this library for using Image Generation. There is one questions I want to ask.

I want to catch the color in input image. For example.. In this image, I want to depict character's various colors in output image. (yellow, red, blue, green, etc..) But, output image is not depicted the color.

I use this code in generating output image.

with torch.no_grad():
        img = resize_image(HWC3(input_image), image_resolution)
        H, W, C = img.shape

        detected_map = np.zeros_like(img, dtype=np.uint8)
        detected_map[np.min(img, axis=2) < 127] = 255

        control = torch.from_numpy(detected_map.copy()).float().cuda() / 255.0
        control = torch.stack([control for _ in range(num_samples)], dim=0)
        control = einops.rearrange(control, 'b h w c -> b c h w').clone()

        if seed == -1:
            seed = random.randint(0, 999999999)
        seed_everything(seed)

        if config.save_memory:
            model.low_vram_shift(is_diffusing=False)

        cond = {"c_concat": [control], "c_crossattn": [model.get_learned_conditioning([extra_prompt + prompt] * num_samples)]}
        un_cond = {"c_concat": None if guess_mode else [control], "c_crossattn": [model.get_learned_conditioning([negative_prompt] * num_samples)]}
        shape = (4, H // 8, W // 8)

        if config.save_memory:
            model.low_vram_shift(is_diffusing=True)

        model.control_scales = [strength * (0.825 ** float(12 - i)) for i in range(13)] if guess_mode else ([strength] * 13)  # Magic number. IDK why. Perhaps because 0.825**12<0.01 but 0.826**12>0.01
        samples, intermediates = ddim_sampler.sample(ddim_steps, num_samples,
                                                        shape, cond, verbose=False, eta=eta,
                                                        unconditional_guidance_scale=scale,
                                                        unconditional_conditioning=un_cond)
        
        if config.save_memory:
            model.low_vram_shift(is_diffusing=False)

        x_samples = model.decode_first_stage(samples)
        x_samples = (einops.rearrange(x_samples, 'b c h w -> b h w c') * 127.5 + 127.5).cpu().numpy().clip(0, 255).astype(np.uint8)

        results = [x_samples[i] for i in range(num_samples)]

How to catch the color in input image and depict in output image? I think that I will fix this question by modifying the code that detects the boundary. right? I'll be waiting for your good opinions.

Thank you.

[fangfchen]

Maybe you can try multi-control with controlnet_scribble + t2i_color in the webui

[DevJunghun]

@fangfchen Thank you for your comment. But, I'm not using webui. only CLI. I will research about how to use t2i_color in CLI env. Thank you!

[lxxie298]

Hello, I also want to achieve this effect. Do you have any good solutions currently?

[geroldmeisinger]

as fangfchen already statet this can be achieved with multi-controlnet scribble + color. diffusers framework has support for multi controlnet. how to implement this with vanilla pytorch I don't know but you can look it up in the CN a1111 extension. another approach is "regional prompting" where you use different prompts for a specific area in the latent space, e.g. upper left corner has "red ear" added in the prompt and and lower left corner has "green leg" added to it.

[fangfchen]

@lxxie298 I later tried another method, using the color map as input for img2img (with around 0.9 denoising weight), which could also achieve a similar effect. I hope it will be helpful.

[lamper-mit]

@lxxie298 I later tried another method, using the color map as input for img2img (with around 0.9 denoising weight), which could also achieve a similar effect. I hope it will be helpful.

Hey fangfchen, how did this work for you? Were the results promising? I'm worried that a color map would mess pickup the white background and cause issues.