Generating AI images with StableDiffusion - Beginner Guide | and with img2vid

[leerlauf]

thanks for the tutorial it was awesome, helped a lot!

just started thinkering myself and had a couple of questions

is it better to have less prompts or the more detailled promtp possible ?

what are the best settings to done down to speed up image generation and mitigating the downgrade of the generated pictures? i want to dablle till i find the characters i 'like.

what's the process used from authors when generating IA to have consistantly the same characters face ?
for exemple when i find a characters i like through the IA generations, how can i make variants of it efficiently to use in a story?

If it's already answered just tell me where i can search thanks !

From my own experience it is far more important to have the correct prompts (the ones that correspond most closely to the training data), rather than having a lot. You still might need a lot of prompts depending on the number of concrete details you need, but for more simple and straightforward tasks it's best to keep things simple - and for really complex stuff you might just need to use photoshop and a subsequent image to image generation.

When it comes to character consistency, your best bet is creating a character Lora. There are already a bunch of free character Loras around that you could use, and with certain styles you can get a consistent character just from prompting (and many checkpoints can already do popular anime/cartoon characters because of their training data), but if you have something very specific in mind you might just need to train a Lora yourself, or commission somebody to do it for you.

 

[idontjudgebro]

From my own experience it is far more important to have the correct prompts (the ones that correspond most closely to the training data), rather than having a lot. You still might need a lot of prompts depending on the number of concrete details you need, but for more simple and straightforward tasks it's best to keep things simple - and for really complex stuff you might just need to use photoshop and a subsequent image to image generation.

When it comes to character consistency, your best bet is creating a character Lora. There are already a bunch of free character Loras around that you could use, and with certain styles you can get a consistent character just from prompting (and many checkpoints can already do popular anime/cartoon characters because of their training data), but if you have something very specific in mind you might just need to train a Lora yourself, or commission somebody to do it for you.

thks for the answers i appreciate it

Is creating a Lora complicated? it seems like it the way you wrote that we can commision someone to do it

 

[leerlauf]

thks for the answers i appreciate it

Is creating a Lora complicated? it seems like it the way you wrote that we can commision someone to do it

I haven't created a Lora myself yet, but from what I have heard it's not too complicated. Might take you some trial and error to get it right. The first step would be to create some good training data - meaning a bunch of different pictures of the character you want in different poses and angles. Use photoshop and img to img to fix these pics up until they look like they depict the same character (I heard you need about 30 for a character Lora - but again - that's must me quoting some guides I read online - haven't done it myself yet).

You can also look around at the commission section and see if you can find an AI-creator who can train a Lora for you, if you don't feel up for it yourself.

 

[idontjudgebro]

I haven't created a Lora myself yet, but from what I have heard it's not too complicated. Might take you some trial and error to get it right. The first step would be to create some good training data - meaning a bunch of different pictures of the character you want in different poses and angles. Use photoshop and img to img to fix these pics up until they look like they depict the same character (I heard you need about 30 for a character Lora - but again - that's must me quoting some guides I read online - haven't done it myself yet).

You can also look around at the commission section and see if you can find an AI-creator who can train a Lora for you, if you don't feel up for it yourself.

thanks for the helpful tips again it really helps

another question,i want to add text messages to my novel but i'mstruggling to adapt the size to the UI

i'm using "yet another phone for renpy"

is there a default config ? furthermore the default textbox in the zipfile on the official git only contains a little sized one so what is writtent just spill out of the box how can i fix that?

 

[Psan2022]

#1 Addition to generating images with Stable Diffusion
(How to make videos from your images)​

So you have been generating images with Stable Diffusion but you want to make short videos from them? I have a little something for you. This method only uses 6GB of VRAM so you could theoretically use weaker Graphics Card.

What you need:
1. Download FramePack ->

You must be registered to see the links

2. Preferably a good graphics card and/or processor is very recommended. But if you do not have one, it should also work as long it has at least 6GB VRAM.

Installation:
1. Downlolad the archive under the Installation section. Unpack the zip folder and put the unpacked folder somewhere with a lot of free space. You need at least 40 GB free space later on!
2. After you have your folder in a desired path klick on the update.bat. Now it will download everything it needs to function. That may take a while since it will download about 30-40 GB of data.
3. After it updated you can close the cmd window and open the run.bat.
4. It will open an browser window. (Similar to Stable Diffusion)


Overview:


UI:
It is very similar to what you know from Stable Diffusion
- In the top left you post your desired picture
- In the top right you have your endproduct
- In Prompt you give a description what you want.
- You can choose a video length of up to 120 seconds (2 minutes)
- In Seed you can put in whatever you want to randomize your output.
- Steps Determine the sampling rate of the generation (the more you choose the longer it will take to generate)
- Distilled CFG Scale is how much you let your pc free to decide the outcome
- MP4 Compression is how the image quality of your video will be

How to generate:
It is quite straight forward.
1. You generate a picture with Stable Diffusion or with another Txt2Img Generator.
2. You put your image in the left upper corner (you can also put in photos and other pictures too).
3. Set the prompt with what is going on in the image.
4. Select a length of the video
5. Press generate.

Here is what I got with just 2 prompts:

The picture:	The Video:
	View attachment 250628_212419_165_9204_37.mp4

It could use some work. But for a quick an dirty way. It does not seem bad at all.
You can find your outputs in:
YourDirectory/framepack_cu126_torch26\webui\outputs

I hope you found that little excursion as interesting as I have.

 

[lordbolton]

copied from Civitai
Wan Self Forcing Rank 16 (Accelerator)

Spoiler: check the spoiler for more info.

You don't have permission to view the spoiler content. Log in or register now.

You must be registered to see the links

I cannot believe this lora works cuts the video generation time to 3 minutes to generate a video, before it was 25 minutes.