Hi there!

Here's a couple of clips from an AI-generated gameshow I'm generating. Currently I'm only aiming to create a TV-style bumper / promo for this content as a complete piece - but with enough interest, I could potentially see it going further.

For the admins (and anyone interested) - all featured persons are fully AI generated. I first created an image of each character using Gemini / Flow, then I created various starting images using the same tools. Afterwards, I animate those images using either Veo or Grok. I then run them through NTSC-RS to give them a little 90s TV authenticity.

Currently I only have two clips, one non-messy featuring "Hazmat Hotties" who would act as stagehands, administering the mess. The other is a clip of two contestants kissing in a vat of gunge.

Please let me know how you think!

Kissing with blue and yellow gunge - https://www.redgifs.com/watch/questionableillinformedbee

"Hazmat Hotties" Scene - https://www.redgifs.com/watch/seriousbustlingpilchard

I really like it! Especially the filter you put over it, that's really effective. The kissing is pretty hot although it's pretty ravenous to the point that the girl on the right looks like she is almost trying to eat the girl on the left. But I love the extra splatter of gunge while she does kiss her.

That first gif? WOW.

I'm about as skeptical of all things AI as you can get. But in terms of realistic *gunginess*, that is far above any of the other AI video I've seen. The aesthetic of it is perfect, too.

Genuinely interested to see where this goes, amazing work.

Thank you! Glad to hear people are enjoying

For anyone interested, from my experiences Google Nano Banana Pro / Nano Banana 2 (for images) and grok (for video) tends to respond well to certain keywords when describing gunge:

- For multiple colours of gunge, described the gunge as "marbled" to get an appealing mixed pattern, instead of it trying to mix the colours (often wrongly) or creating gunge with mostly one colour and a few small splodges of the other colour.

- Especially in grok, describing the gunge as "very thick" helps keep the gunge from looking watery.

- If creating an overhead pour video, be very explicit with how you want it to look. This includes the pour ("a huge endless column of very thick gunge" is a good starting point), how it hits the characters ("the deluge pours / splatters / rains down on the character" all produce different effects), and how it covers them (IE, "the gunge covers their face and hair completely", "the gunge pours down their bodies in thick waves", etc)

- Treat your prompt as a linear script. Describe your actions in sequence and use words like "before" and "then" with directions to have multiple actions happen in order.

- Use camera direction! An "amateurish handheld camera" will produce results like the above, along with directions like "the camera moves in to close up".

- If you're having trouble keeping your actors visually consistent (disappearing glasses, changing outfits, etc) describe them at the start of the prompt, making sure to include any details that were generated inconsistently last time. AI only knows what you write and your starting image(s), so make sure to give any details that aren't immediately visible in the prompt!

- If the gunge stops pouring and you want a longer pour, at the end add the phrase "the gunge continues to pour throughout the entire scene".

- if you're going for a 90s TV look with a video, crop the starting image to a 4:3 resolution beforehand. This yields dramatically better results as the aspect ratio provides more information on the type of training data the AI will take inspiration from than you'd think!

I should also note, as well as having an affinity for it, one of the main reasons I use the 90s gameshow and handheld camera aesthetics in my generations is that it helps to hide AI jankiness, and when a shot is too clean / perfect it looks very obviously fake.

For anyone who likes experimenting with AI, I hope these tips come in handy! If anyone is interested I am happy to provide a set of pre-written prompts (for both image generation and video generation) to get people started