Video yükleniyor...
Video Yüklenemedi
am testing all of the leading video generation models using the same prompt in thread. model: seedance 2.5 on fal music: Suno editing: ffmpeg using Codex
260,106 görüntüleme • 10 gün önce •via X (Twitter)
55 Yorum

seedance 2.5 prompt: """ This is messy, fast-cut 1980s home video shot on a VHS camcorder in a "plot" — a cluster of rented rooms around a shared dirt courtyard — in Nairobi's Jericho estate, Kenya. All people are Kenyan, and any speech, shouting, or chatter heard is in Swahili or Sheng, with the natural rhythm and intonation of East African Swahili speakers. Cuts happen very frequently, around every 1.5 seconds. The camera jumps quickly between lots of different shots: different angles of the mabati (corrugated iron) roofs and doorways, people outside, the dirt courtyard, clotheslines strung between buildings, a parked Peugeot 504 or VW Kombi, jerry cans by the communal water tap, and random moments happening around the compound. Each person has exactly two hands and anatomically correct arms; actions like opening a tap or pinning clothes to a line are done simply and naturally with two hands, clothespins pushed down onto the line from above. The editing feels spontaneous and slightly chaotic, as if the person filming keeps changing their mind about what to show. The camera is extremely shaky and handheld, with constant movement and loose framing. The footage has the distinctive texture and color of real 1980s VHS video. Natural sound only, recorded with the built-in microphone. The result must feel like genuine, fast and chaotic home video from someone randomly filming life in a Nairobi housing estate in the 1980s. """

suno prompt for music: """ Authentic 1980s Kenyan benga and East African guitar pop, pure instrumental, lively Nairobi estate courtyard party energy, 128 BPM in C major, 4/4 driving shuffle groove, Build the arrangement around two interlocking clean electric guitars: bright chiming rhythm guitar with tight offbeat accents, and a melodic lead guitar trading concise call-and-response phrases, Add occasional fast soukous-style guitar flourishes without losing the benga pulse, Use a rolling, highly melodic electric bassline that drives the groove, plus a tight drum kit led by crisp hi-hat shuffle, snappy rimshot-style backbeat, and lively kick pattern, Keep the performance communal, joyful, slightly loose, and danceable, Warm saturated analog 1980s cassette production with narrow stereo image, gentle compression, modest spring-like room ambience, cheap funk/reggae/rock warmth, tape hiss, scratchy charm, softened transients, and subtle wow-and-flutter, No vocals, no singing, no spoken parts. """

@fal @suno that vhs courtyard chaos feels stolen from a real 80s tape. how close is seedance getting?

@fal @suno i mean this was really good for one shot.

@fal @suno same prompt across models is the only fair test. I'm watching which one keeps motion without melting. Seedance winning for you so far or still early?

@fal @suno winning for now but early to tell. testing gemini omni 1.1 flash now.

@fal @suno Y'all that laugh your African parents for not knowing the difference between AI videos and real videos, your time is up! 😭😭😂

@fal @suno If I didn’t see caption I wouldn’t think it’s ai generated Damn I pity my older days

@fal @suno Ohhhh this is scaringly goood! WTF

@fal @suno yes, we're there.

Os modelos não se comportam da mesma forma. Colocar o mesmo prompt é injusto, os modelos tendem a trazer resultados até parecidos, mas se você quiser extrair o melhor de cada um, é preciso fazer o prompt conforme cada modelo responde. Tente um prompt com acting e diálogo que você verá porque reaproveitar prompt não é o melhor teste.

@fal @suno this is not ai

@fal @suno it is

@fal @suno what are you scoring beyond the visuals — prompt adherence, temporal consistency, or how much ffmpeg cleanup each clip needs?

@fal @suno prompt adherence and cosistency.

@fal @suno Damn that's good Tooooo good

@fal @suno yes! seedance did cook a good model tbh

@fal @suno This is the best

@fal @suno the woman just dey do rubbish say she wan dry cloth yall need to stop this rubbish

@fal @suno 😂😂😂

@fal @suno You used a Reference Image right ?

@fal @suno nope! not on this first one.

@fal @suno This is insanely good! How many credits did you use?

@fal @suno From @Script_clips, the cost stands out here this is less than $2 when Seedance is about $10 per run.

@fal @suno Same prompt across all models is the only honest way to compare. Most people cherry pick the best take from their favorite tool and call it a benchmark.

@fal @suno omo

@fal @suno Nice!

@fal @suno

@fal @suno impressive stuff. But is that woman talking sheng?

@fal @suno hiyo ata sijui ni language gani? 😂

@fal @suno No way this is ai

@fal @suno it's wild

@fal @suno True.

@fal @suno looks really good! I think seedance 2.5 is really the best model at the moment, but it is also pretty expensive. also love minimax h3 for its speed and efficiency!

@fal @suno Cool stuff

@fal @suno I could spot some AI flaws in it.

@fal @suno yes, there'll awlays be signs but crazy improvement from Will Smith and spaghetti era

@fal @suno True

@fal @suno This is mind blowing

@fal @suno A Seedance 2.5 workflow like this is more reliable when the process does not end at a one-off render; once the idea is set, Atlas Cloud keeps the next Seedance 2.5 task connected to the inputs that shaped it.

@fal @suno Interesting.

@fal @suno This is so good 👍

@fal @suno have you tried @Palmier_io ? you can generate seedance video, music, and edit all in one place

@fal @suno @Palmier_io happy to try it. free trial?

@fal @suno @Palmier_io DM'd you!

@fal @suno Bro, this is incredible!

@fal @suno Looks amazing. How does the Swahili sound?

@fal @suno the swahili is wrong haha

@fal @suno Ah…. Probably easily fixable though and still pretty amazing for a one shot. Very cool experiment 🙏

@fal @suno yes indeed.

@fal @suno uhn?? 😳

@fal @suno This kind of side-by-side testing is way more useful than any of the marketing pages for these models. What's your read on seedance 2.5 so far

@fal @suno huh?

@fal @suno It remembers me my childhood in Cameron.

@anto_ty @fal @suno Dope dope dope dope dope…….. allow me to play with the prompt
