[GET THE OLD SAMPLING METHOD BACK]:

[37OMKAR]

[GET THE OLD SAMPLING METHOD BACK]:

[AlexysLovesLexxie]

I mean, I have everything working as intended again. It's not that much different from how it was. The results I can get (even screwing up and forgetting to switch from DPM++ 2M to DPM++ 2M Karras) are way better than I could get out of comfy with the same model, resolution, steps, and sampler in Comfy. I do agree that the new system isn't great, though.

…

On Sat., Apr. 27, 2024, 10:44 p.m. 37OMKAR, ***@***.***> wrote: having just tried Comfy (and hated it) I feel that these changes seem to be aiming to "make A1111 just like comfy". Unless I'm wrong, and there's actually a good reason for these changes? comfy is hard so people prefer a1111, but some changes are not understandable to all genral users, as all are not technical, yes we can understand some pro need it, but look out for the percantage of users need it, it wil very less, majortiy are non tecnical people — Reply to this email directly, view it on GitHub <#15644 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A6I5UHJUMTLVJXD4XXP6S3TY7SECNAVCNFSM6AAAAABG4F56TKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAOBRGM2DCNRRGQ> . You are receiving this because you commented.Message ID: ***@***.***>

[AG-w]

just change scheduler, it's right next to the sampler

[37OMKAR]

just change scheduler, it's right next to the sampler

yes i see that, but its not creating images as it was, now if i need ony euler, i have to select someting , there are people who are not that much tech savy using the A1111, and this updated is bit hard, can we get old verion back, like a extnsion, it will be massive help to people like us who cant get all this jargon and tecnical stuff, being a agriculture graduate, its too hard to google every stuff and get things working, and once we get work flow going there is a updated, we request just to enable us to get that old version or sampling as extension so we can use it as it was, and pros can use it seprated, thanks

[AG-w]

you can download older version and replace the current one

[AlexysLovesLexxie]

having just tried Comfy (and hated it) I feel that these changes seem to be aiming to "make A1111 just like comfy". Unless I'm wrong, and there's actually a good reason for these changes?

[37OMKAR]

having just tried Comfy (and hated it) I feel that these changes seem to be aiming to "make A1111 just like comfy". Unless I'm wrong, and there's actually a good reason for these changes?

comfy is hard so people prefer a1111, but some changes are not understandable to all genral users, as all are not technical, yes we can understand some pro need it, but look out for the percantage of users need it, it wil very less, majortiy are non tecnical people

[37OMKAR]

if we want complexity as comfy then people will use that node base genration, and to be clear i use blender, and i know majority of belnder user dont use geomety nodes as this is node base and it gives a vibes of complication and lot of people are actully afrid, like they get anixty even looking at it, same is with me, node base is best but its a logic base and majority dont get the logic, so majority dont use comfy, so its my personal view that keeping it clean and easy will bring more users, there must be pro version but that must be made avalable by way of a exrension or tick

[Soulreaver90]

That brought up a question I had the other day, what scheduler does Euler/A use? Do we leave it on automatic? Is it none (not an option) is it exponential? It makes sense for DPM, but not for others. The other issues is automatic, what does it mean? Does it choose a type at random? Is it the best type for that scheduler?

[37OMKAR]

This has now opened a Pandora box, no one like us actually know what it dose, it's just you select do image and see what combination fits perfect for a model, as they all behave differently for different models as per I know, so ya it's just a drop down you select till you get your desire resut that is what I know of it and I use it that way

…

On Sun, 28 Apr, 2024, 2:04 pm Soulreaver90, ***@***.***> wrote: That brought up a question I had the other day, what scheduler does Euler/A use? Do we leave it on automatic? Is it none (not an option) is it exponential? It makes sense for DPM, but not for others. The other issues is automatic, what does it mean? Does it choose a type at random? Is it the best type for that scheduler? — Reply to this email directly, view it on GitHub <#15644 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A2GBX5QB7DVXDNV64S3PS63Y7SYCFAVCNFSM6AAAAABG4F56TKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAOBRGM4TCOJVGY> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[Soulreaver90]

This has now opened a Pandora box, no one like us actually know what it dose, it's just you select do image and see what combination fits perfect for a model, as they all behave differently for different models as per I know, so ya it's just a drop down you select till you get your desire resut that is what I know of it and I use it that way
…
On Sun, 28 Apr, 2024, 2:04 pm Soulreaver90, @.> wrote: That brought up a question I had the other day, what scheduler does Euler/A use? Do we leave it on automatic? Is it none (not an option) is it exponential? It makes sense for DPM, but not for others. The other issues is automatic, what does it mean? Does it choose a type at random? Is it the best type for that scheduler? — Reply to this email directly, view it on GitHub <#15644 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/A2GBX5QB7DVXDNV64S3PS63Y7SYCFAVCNFSM6AAAAABG4F56TKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAOBRGM4TCOJVGY . You are receiving this because you authored the thread.Message ID: @.>

I don’t think the old method needs to come back, I just think the new one needs to be further refined as it’s currently half baked. We need a default or none option for ones like Eular, LCM, etc. Automatic needs to tell us, if it selects one at random, which one. Maybe the best option is to offer a choice between simple and advanced. Where simple is basically the old method with a list of the most common samplers/types, while advanced is the current method with all at your disposal. Honestly A1111 is getting more bloated with each release, it might be time to look into a simple and advanced version to tailor both new and seasoned prompters. I think half the reported issues could be avoided if newbies just stick to select and generate.

[AG-w]

I don't know how is this "hard" at all, all it did is separated sampler and scheduler
in fact, it allow people to pair different sampler and scheduler without messing with code, it's more convenient
I actually wonder why this hasn't been implemented sooner

also all this "not all people is tech savvy", serious? I recommend people to use fooocus if they think this way

issues could be avoided if newbies just stick to select and generate.

yea, fooocus is for them

Automatic needs to tell us,

there's nothing special about automatic, check sd_samplers_kdiffusion.py, it just use default setting

[37OMKAR]

I don't know how is this "hard" at all, all it did is separated sampler and scheduler in fact, it allow people to pair different sampler and scheduler without messing with code, it's more convenient I actually wonder why this hasn't been implemented sooner

also all this "not all people is tech savvy", serious? I recommend people to use fooocus if they think this way

issues could be avoided if newbies just stick to select and generate.

yea, fooocus is for them

Automatic needs to tell us,

there's nothing special about automatic, check sd_samplers_kdiffusion.py, it just use default setting

i have said REQUEST, no where we are saying its must be done, and if we have a point of view we must be allowed to put it, i never said remove it, i said, if possible, provide that old as a someting as extenstion, we are happy that dev are putting efforts its just a prespetive, and i think we can put our views here

[missionfloyd]

That brought up a question I had the other day, what scheduler does Euler/A use? Do we leave it on automatic? Is it none (not an option) is it exponential? It makes sense for DPM, but not for others. The other issues is automatic, what does it mean? Does it choose a type at random? Is it the best type for that scheduler?

Automatic chooses the default from this list. For samplers without one listed, (like Euler), you can leave it on Automatic and it'll be the same as it was before, or try others if you want to.

stable-diffusion-webui/modules/sd_samplers_kdiffusion.py

Lines 11 to 27 in 1c0a0c4

	samplers_k_diffusion = [
	('DPM++ 2M', 'sample_dpmpp_2m', ['k_dpmpp_2m'], {'scheduler': 'karras'}),
	('DPM++ SDE', 'sample_dpmpp_sde', ['k_dpmpp_sde'], {'scheduler': 'karras', "second_order": True, "brownian_noise": True}),
	('DPM++ 2M SDE', 'sample_dpmpp_2m_sde', ['k_dpmpp_2m_sde'], {'scheduler': 'exponential', "brownian_noise": True}),
	('DPM++ 2M SDE Heun', 'sample_dpmpp_2m_sde', ['k_dpmpp_2m_sde_heun'], {'scheduler': 'exponential', "brownian_noise": True, "solver_type": "heun"}),
	('DPM++ 2S a', 'sample_dpmpp_2s_ancestral', ['k_dpmpp_2s_a'], {'scheduler': 'karras', "uses_ensd": True, "second_order": True}),
	('DPM++ 3M SDE', 'sample_dpmpp_3m_sde', ['k_dpmpp_3m_sde'], {'scheduler': 'exponential', 'discard_next_to_last_sigma': True, "brownian_noise": True}),
	('Euler a', 'sample_euler_ancestral', ['k_euler_a', 'k_euler_ancestral'], {"uses_ensd": True}),
	('Euler', 'sample_euler', ['k_euler'], {}),
	('LMS', 'sample_lms', ['k_lms'], {}),
	('Heun', 'sample_heun', ['k_heun'], {"second_order": True}),
	('DPM2', 'sample_dpm_2', ['k_dpm_2'], {'scheduler': 'karras', 'discard_next_to_last_sigma': True, "second_order": True}),
	('DPM2 a', 'sample_dpm_2_ancestral', ['k_dpm_2_a'], {'scheduler': 'karras', 'discard_next_to_last_sigma': True, "uses_ensd": True, "second_order": True}),
	('DPM fast', 'sample_dpm_fast', ['k_dpm_fast'], {"uses_ensd": True}),
	('DPM adaptive', 'sample_dpm_adaptive', ['k_dpm_ad'], {"uses_ensd": True}),
	('Restart', sd_samplers_extra.restart_sampler, ['restart'], {'scheduler': 'karras', "second_order": True}),
	]

Samplers that had Karras or Exponential in their names before are now just spilt between the two boxes. E.g., for DPM++ 2M Karras, select DPM++ 2M and Karras.

[AlexysLovesLexxie]

On Mon., Apr. 29, 2024, 5:59 p.m. missionfloyd, ***@***.***> wrote: Samplers that had Karras or Exponential in their names before are now just spilt between the two boxes. E.g., for DPM++ 2M Karras, select DPM++ 2M and Karras.

I think the problem.mis that most people didn't memorize which variants each sampler had. I know that 2m and 3m SDE both had Karras, but I couldn't tell you anything about the others as I personally never used them. But I'm also not a "not tech savvy" user, and have ways of figuring this out, including using my 1.7.0 install to make notes. This is why I make parallel installs and never just upgrade. 4TB SSDs are dirt cheap these days.

…

[37OMKAR]

i just went back to 17 version and that more easy

[Confuzu]

Another problem is with the XYZ Plot. With the old system, the Sampler needed only one value to try every sampler to find the best for the prompt. The other two could be used for checkpoints, clip skip or any other.
With the new system sampler needs two X and Y to run all sampler.