I'm joining the Realistic Challenge inside @Daniel Riley AI VideoBootcamp, so I thought I’d share each prompt I'll use for it, here with you every day too. Yesterday was Day 1: Portraits. I’ll drop the exact prompt in the first comment. The first 3 images were made with Recraft 4.1, and the last 5 with ChatGPT Image 2. For realistic portraits, I’d personally recommend Recraft, but Seedream 5 Pro and ChatGPT Image 2 are still really good options. This time ChatGPT came out a little more dreamy, so I ended up choosing image 1 for Day 1. Which one would you have chosen? :) It was late at night when I picked it... looking at them again now, I’m not so sure anymore 😂 I like 1, 3 and 5.

A woman arrives by a vintage sports car on a sun-drenched LA hillside, steps out, walks away, then appears across a poolside estate, changing wardrobes, cinematic live-action, static camera, hard cuts between shots. This video clip-test, is inspired by Mango's and Hailey Bieber's video featuring their summer collection 2026, which I reverse engineered on this class's VIP Vault, and then I provided parts of the script to the agent, which created a reference sheet for the model and two 2x2 storyboards on 16:9, and then generated the whole campaign on 9:16. Originally the video would have been 27sc with the 3 storyboards, but I chose to do it 15sc The whole process is so effortless and so fast. The whole video is one take. At the pool there are two pair of shoes:))) This short AI video clip is generated exclusively with the Ai Agent>Ad campaign on Runwayml with Seedance 2.0 https://www.instagram.com/p/DcwO99dg8VI/

Hey everyone, Seedance 2.5 is officially on the way from ByteDance. The big upgrade is 30-second video generation in a single pass, alongside stronger reference control, better editing, smoother motion and more production-focused tools. Higgsfield, OpenArt, Dreamina and fal.ai has already announced upcoming access with probably all the rest coming just as fast when they release th API key to the public. But I’ve also seen some places saying that CapCut already has it. (Could be location based) The model is expected to support up to 50 image, video and audio references in total, with reported 4K output. But again the best thing is 30 seconds long output !!😮‍💨🔥 (Don’t expect it to be cheap tho 🫠) but imagine a seedance 2.5 ‘lite’ :) that would be the bomb just for continuity alone on easier stuff like animation .. etc It looks like the rollout is starting now. I’ll keep watching the access situation and share more once it’s properly live for testing. 👀 https://seed.bytedance.com/en/seedance2_5 Meanwhile below are some examples to get you excited

Hey everyone, quick thing for everyone to consider. When you make a video, you need to make sure you start from a great first image (that’s why it’s called the HERO shot). But once you have that locked as your north start.. you need to stop looking at your best frame and get complacent with how good that is. And start looking for the weakest one, because that’s usually the frame that gives the whole thing away. Sometimes it’s a weird texture, sometimes the product label breaks, sometimes the face is a lot shinier all of a sudden, sometimes the motion feels gamified, sometimes the last 2 seconds of the generation feels off.. CUT IT! That one part.. gives EVERYTHING AWAY. All the hard work you’ve put into everything else in the project. It’s not even about hiding its AI, it’s about having a high level consistency throughout. People don’t really care if it’s AI anymore, as long as it good. They care if it’s Slop, or inconsistent. Most people won’t always know exactly what’s wrong, but they’ll feel it. So before calling something done, I usually go through it and ask myself: what is the weakest frame that gives it away as a low effort video? - The one that it’s not at the level of the Hero frame we started with. That’s the one to consider if it’s worth keeping in or cut it out/rework it. If you are working with 1 image - 1 generation at a time, than that should happen before turning stills into video, because if the still/image already has problems/ doesn’t look as good as the rest, animation usually makes them worse. Good portions of the generations always steal our focus.. but don’t let bad parts slip in just because you like 20% of it.. that’s what will make you stand out from everyone else. This is a simple yet hard rule to live by. But I advise everyone to try and follow the weakest frame :) 🥂