Add option to img2imgalt.py to use sigma adjustment instead of original method for #736

This commit is contained in:
Martin Cairns 2022-09-26 21:13:23 +01:00 committed by AUTOMATIC1111
parent c74becca23
commit 258a2d4f06

View file

@ -59,7 +59,55 @@ def find_noise_for_image(p, cond, uncond, cfg_scale, steps):
return x / x.std()
Cached = namedtuple("Cached", ["noise", "cfg_scale", "steps", "latent", "original_prompt", "original_negative_prompt"])
Cached = namedtuple("Cached", ["noise", "cfg_scale", "steps", "latent", "original_prompt", "original_negative_prompt", "sigma_adjustment"])
# Based on changes suggested by briansemrau in https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/736
def find_noise_for_image_sigma_adjustment(p, cond, uncond, cfg_scale, steps):
x = p.init_latent
s_in = x.new_ones([x.shape[0]])
dnw = K.external.CompVisDenoiser(shared.sd_model)
sigmas = dnw.get_sigmas(steps).flip(0)
shared.state.sampling_steps = steps
for i in trange(1, len(sigmas)):
shared.state.sampling_step += 1
x_in = torch.cat([x] * 2)
sigma_in = torch.cat([sigmas[i - 1] * s_in] * 2)
cond_in = torch.cat([uncond, cond])
c_out, c_in = [K.utils.append_dims(k, x_in.ndim) for k in dnw.get_scalings(sigma_in)]
if i == 1:
t = dnw.sigma_to_t(torch.cat([sigmas[i] * s_in] * 2))
else:
t = dnw.sigma_to_t(sigma_in)
eps = shared.sd_model.apply_model(x_in * c_in, t, cond=cond_in)
denoised_uncond, denoised_cond = (x_in + eps * c_out).chunk(2)
denoised = denoised_uncond + (denoised_cond - denoised_uncond) * cfg_scale
if i == 1:
d = (x - denoised) / (2 * sigmas[i])
else:
d = (x - denoised) / sigmas[i - 1]
dt = sigmas[i] - sigmas[i - 1]
x = x + d * dt
sd_samplers.store_latent(x)
# This shouldn't be necessary, but solved some VRAM issues
del x_in, sigma_in, cond_in, c_out, c_in, t,
del eps, denoised_uncond, denoised_cond, denoised, d, dt
shared.state.nextjob()
return x / sigmas[-1]
class Script(scripts.Script):
@ -78,9 +126,10 @@ class Script(scripts.Script):
cfg = gr.Slider(label="Decode CFG scale", minimum=0.0, maximum=15.0, step=0.1, value=1.0)
st = gr.Slider(label="Decode steps", minimum=1, maximum=150, step=1, value=50)
randomness = gr.Slider(label="Randomness", minimum=0.0, maximum=1.0, step=0.01, value=0.0)
return [original_prompt, original_negative_prompt, cfg, st, randomness]
sigma_adjustment = gr.Checkbox(label="Sigma adjustment for finding noise for image", value=False)
return [original_prompt, original_negative_prompt, cfg, st, randomness, sigma_adjustment]
def run(self, p, original_prompt, original_negative_prompt, cfg, st, randomness):
def run(self, p, original_prompt, original_negative_prompt, cfg, st, randomness, sigma_adjustment):
p.batch_size = 1
p.batch_count = 1
@ -88,7 +137,10 @@ class Script(scripts.Script):
def sample_extra(conditioning, unconditional_conditioning, seeds, subseeds, subseed_strength):
lat = (p.init_latent.cpu().numpy() * 10).astype(int)
same_params = self.cache is not None and self.cache.cfg_scale == cfg and self.cache.steps == st and self.cache.original_prompt == original_prompt and self.cache.original_negative_prompt == original_negative_prompt
same_params = self.cache is not None and self.cache.cfg_scale == cfg and self.cache.steps == st \
and self.cache.original_prompt == original_prompt \
and self.cache.original_negative_prompt == original_negative_prompt \
and self.cache.sigma_adjustment == sigma_adjustment
same_everything = same_params and self.cache.latent.shape == lat.shape and np.abs(self.cache.latent-lat).sum() < 100
if same_everything:
@ -97,8 +149,11 @@ class Script(scripts.Script):
shared.state.job_count += 1
cond = p.sd_model.get_learned_conditioning(p.batch_size * [original_prompt])
uncond = p.sd_model.get_learned_conditioning(p.batch_size * [original_negative_prompt])
rec_noise = find_noise_for_image(p, cond, uncond, cfg, st)
self.cache = Cached(rec_noise, cfg, st, lat, original_prompt, original_negative_prompt)
if sigma_adjustment:
rec_noise = find_noise_for_image_sigma_adjustment(p, cond, uncond, cfg, st)
else:
rec_noise = find_noise_for_image(p, cond, uncond, cfg, st)
self.cache = Cached(rec_noise, cfg, st, lat, original_prompt, original_negative_prompt, sigma_adjustment)
rand_noise = processing.create_random_tensors(p.init_latent.shape[1:], [p.seed + x + 1 for x in range(p.init_latent.shape[0])])
@ -121,6 +176,7 @@ class Script(scripts.Script):
p.extra_generation_params["Decode CFG scale"] = cfg
p.extra_generation_params["Decode steps"] = st
p.extra_generation_params["Randomness"] = randomness
p.extra_generation_params["Sigma Adjustment"] = sigma_adjustment
processed = processing.process_images(p)