I might make a full post on this in future. I use image to video exclusively. I'm fortunate in that my tastes are quite repetitive, and I can re-use the same prompt with different images to a good degree of success.

I set the relevance to 0.9. I want it to follow my prompt closely. I only prompt in pro mode, as there is no way of recreating a result from standard mode. I use version 1.5, which doesn't allow prompt extensions. If you've seen my stuff, you'll know I like gunge being poured overhead, with a lively reaction from the subject. Here are some things I've found:

- If I want the gunge to pour from above, I need to have to the gunge already pouring in the prompt image. Otherwise, there is a good chance it will materialise on the subject from thin air. There is an optimal 'thickness' of the gunge pour that leads to the best results.
- the gunge will follow the properties of the gunge in the starting image, but I can reinforce one aspect of that in the text prompt (e.g., translucent, or chunky, or sloppy). The viscosity tends to be determined by the image prompt.
- if I want the gunge to stick to the subject, I need to put this in the text prompt (e.g., "the gunge is sticky and sticks to her fair, face, body and legs as it flows down"). This works well, however, weirdly hydrophobic hair can still happen.
- the reaction described in the text prompt should be compatible with the image. In general, the more synergy between the text and the image, the better the result will be.
- synonyms for gunge are all possible, 'sludge' seems to work well for me.
- the more complex the image prompt or text prompt, the more there is to go wrong. The best results are from a single subject with no background people. The more realistic the image prompt, the better the result will be.
- if something in the periphery of the video ruins an otherwise good result, I can crop it out of the video using editing software (making sure to use the best interpolation option).

I'll buy some credits to experiment, then use the daily allowance to make 2 clips a day until it refreshes or I want to buy more. I hope this is helpful to you the next time you get credits.

Thereald, thank you for the pointers. I have had a problem with mess 'sticking' and I will certainly try what you suggested next time. I've been buying extra credits and although they only go down by 10 at a time, it seems that they go down WAY too fast. Smarter prompting (using some of your suggestions) will likely help me have fewer wasted results and more keepers. So thanks again, your comments are appreciated!

Some short video GIFS

Nicely done!!!

I took some of my better results, tweaked the text prompt and regenerated in professional mode with version 1.5. Initially the 35 credits seems exorbitant, until I realized I'd run the same prompt in standard mode four times for 40 credits, and none of the results were worth it.
Version 1.5 however does not support the motion brush function that enables you to direct the movement of assets, so the prompt must be pristine. It does produce spectacular fidelity though.

Thanks To thereald for advice on fine tuning the settings, really helpful and appreciated.

I did also try to extend a professional clip by capturing the final frame to utilize as an image to generate the next one, not ideal.

Kling also seems to devour bandwidth - be careful if you're on a mobile device or hot-spotting.

That one is amazing

Man... that's impressive. We're like two or three big developments from crazy stuff.

Episode 14 of STB is up.

https://youtu.be/sjekiWTFzSw?si=0L8EZNVYidVgwqge

This is the penultimate episode of the first season, the final episode will be out next Saturday. Suggestions will be considered for season 2, which should be sometime next year.

An original AI video for the game show, Custard or Mustard . . . https://youtu.be/w64K8Sw-ImY

We're getting closer to the crazy stuff, but I predict the last steps will be hard ... but hey ... pretty much two years ago I generated my first AI image (see below) with Dall-E and using one of my own images as a prompt ... and today I can do minute long 4K movies ...



https://www.youtube.com/watch?v=h0Ws8fx5iJA

These are absolutely incredible. If you don't mind me asking, how do you do these? What software /prompts did you use?

My last video based on a vote, which model to mess up ... I decided to mess them all up anyways ... been experiment with different settings, outfits, hairstyles etc.

https://www.youtube.com/watch?v=DjjiNVgPj9g

I use a varying set of tools, depending on where I want to end up. There's non one-stop-shop as of now to achieve high quality results. The tools also evlove over time. But in any way, none of them are free.

Just put this up - season 2 preview:

https://www.youtube.com/shorts/ba3DcMb4MDg

Upping the AI video Game a bit ...
https://www.youtube.com/watch?v=Doh4pCUwPHY

S2E2:

https://www.youtube.com/watch?v=-lI6liNQ-Q4&t=2s

No-one will see this because of the bump filter

At least the video shows up in the forum video archive! Great work anyway!

Here's the deal ... you get a perfect makeover, but you have to be featured in a WamPhoto AI video ... https://www.youtube.com/watch?v=Hgnbj8Ov_2Q

I finally compiled some of my best Kling clips into a 3 min video - each clip is 5 seconds so there's 36 unique clips in total. It's a futuristic game-show theme with swimsuits and slime aplenty in HD quality.

I have a lot more clips with the same theme, so I may compile a longer video in the near future.

Enjoy!

https://umd.net/videos/ai-slime-compilation

My second compilation completed, stay tuned to the end for an intriguing new concept.

https://umd.net/videos/ai-compilation-video