vladmandic
commented
1 year ago native format in memory is not png, its just raw - so converting to either png or jpeg is about the same.
some params you're sending make no sense, for example do_not_save_samples is not a valid request param, its calculated from save_images. and you're sending both sampler_name and sampler_index - its one or the other.
plus Euler is disabled by default (because its ooooold and pointless compared to any newer ones), so its likely causing a fallback to some other sampler - and that may cause some delay.
first thing, i'd reduce the json payload to absolute minimum and use values that are known to be good and see if that helps.
blacklig
Michaelvirga
Issue Description
Hi!
First of all, thanks vladmandic for great fork of Automatic1111. Your build is generally better and I think also faster than A1111.
And now if anyone can help me with my issue:
In one task I cannot get anywhere close in performance as from A1111, it is getting response from API endpoint img2img.In A1111 I am getting response back like "instantly", I would say 100-200ms and response is getting back to me from server, on this fork it is like 1 second hang there before I get response and I really dont know why and spent some time on it and still cannot find you why..
I describe it very primitively - In one terminal window I have running instance of Stable Diffusion in this implementation Automatic and and on another window I have postman with payload. Once I send it, I see in terminal like that animation of image being generated and once it finishes (in like second and half on 3090), then it adds also another second I would say before postman gets result. (actually in script it is same result, so nothing wrong with postman). But when I do exactly the same thing with Automatic1111 instance, sending same prompt, generally on the same server settings etc, I am getting response from server almost instantly. Like once that animation shows it is finished, like after 100-200ms I am getting response back to postman or into script.
So somehow I am getting like +1s on every response from this fork and I really dont know why. Actually image generation is bit faster on this fork than on A1111, probably due to properly installed libraries etc but this issue totally kills it.
Only difference I would say is in this fork there is default working format .jpeg so response gets back in jpeg and probably there is conversion from PNG but no way it could get 1s on such small images I am using.. so even though A1111 is sending back 200kb png it is way faster than automatic with 50 kb jpeg.
Endpoint: http://127.0.0.1:7860/sdapi/v1/img2img
Payload I am sending is this:
{ "init_images": [ "/9j/4AAQSkZJRgABAQAA...." ], "resize_mode": 0, "denoising_strength": 0.75, "image_cfg_scale": 7, "inpainting_fill": 0, "inpaint_full_res": true, "inpaint_full_res_padding": 0, "inpainting_mask_invert": 0, "initial_noise_multiplier": 0, "prompt": "beba", "styles": [ "" ], "seed": -1, "subseed": -1, "subseed_strength": 0, "seed_resize_from_h": -1, "seed_resize_from_w": -1, "sampler_name": "Euler", "batch_size": 1, "n_iter": 1, "steps": 50, "cfg_scale": 7, "width": 512, "height": 512, "restore_faces": false, "tiling": false, "do_not_save_samples": false, "do_not_save_grid": false, "negative_prompt": "", "eta": 0, "s_min_uncond": 0, "s_churn": 0, "s_tmax": 0, "s_tmin": 0, "s_noise": 1, "override_settings": {}, "override_settings_restore_afterwards": true, "script_args": [], "sampler_index": "Euler", "include_init_images": false, "script_name": "", "send_images": true, "save_images": false, "alwayson_scripts": {} }
and bonus question: is it also possible to send Image as link on local drive or something like that instead of base64? I saw in webui using WS this actually is exactly like this but somehow cannot get it work via this api.
Version Platform Description
Ubuntu, RTX 3090