Every change to the bot, newest first. The command guide has the full syntax.
Ask the bot itself: /ask what's new? answers from this
page, and /ask changelog: anything about videos? narrows it down.
/chat <message> is a conversation with the AI that remembers what was said. Everyone in a group shares one conversation and it knows who said what. Reply to one of its answers to keep going without a command, or just type in a private chat with the bot. web: makes it look something up first, photos work, and /newchat starts over./gifgen, /animate, /vidgen, /vanimate, /swap, /musicgen and /memeit (and their short forms) answer with a one-line notice instead of running./ask, /joke and /lunch run on a new AI. It is quick, reads photos, and still searches the web when a question needs it, but it is more buttoned-up than the old one and may turn down the rudest requests./imagegen, /edit, /enhance and /merge still work with gemini: and gpt:. The other image model prefixes (flux:, flux2:, dev:, qwen:, flash:, krea:, zimage:) are off for now, and a declined picture is no longer retried on a backup model./weather, /request and the emoticon commands work as before./findimage pepe or /images wolverine finds them. Real people are never named./ask no longer plays along with requests for its API keys, tokens, passwords or config: it has none, and anything it "showed" before was made up on the spot. It now says so instead of inventing one, and any key-looking string in an answer is redacted./ask what's new? (or /ask changelog:) tells you what changed recently, and /help links here too./memeit stays inside its own group: meme videos forwarded in from another chat, or re-posted, no longer count as things people said here./memeit overhaul: it now writes one short comedy sketch (one gag, two people at most, a quoted line and a punchline) instead of a crowded scene, and every run rolls a random video genre: sitcom, nature doc, security cam, telenovela, silent film, news report, infomercial, reality tv, claymation, anime, found footage, action movie, video game, noir, western, cooking show, sports broadcast. /memeit sitcom picks one, /memeit random rolls, and the caption shows which genre you got. The everything-is-neon look is gone./findimage and /images find GIFs by what is in them. GIFs saved earlier without tags were filled in./ask knows the bot: ask it how to do something with the bot ("how do I make a video with sound?", "what can you do?") and it answers from the command guide with the exact command and an example. bot: or help: forces that.gpt: image model moved to ChatGPT's newer, faster model at high quality, for /imagegen, /edit, /merge, and the /enhance fallback./gifgen, /vidgen, or /vanimate render is now 15 seconds (was 2 minutes). Do nothing and it renders the expansion./request <idea> (/req): file a feature request. It goes on a list the maintainer gets as a prioritized digest; /requests shows your open ones./weather <zip> and /lunch <zip> [anything]: weather by US zip code, and real local lunch spots (Yelp included, every pick backed by a source link). The bot remembers your last zip./swap: put the character from a photo into a video. Reply to the video with a photo captioned /swap, or the other way round; GIFs work as the video too./vidgen and /vanimate render about twice as fast at the same quality, and /animate about a quarter faster./gifgen, /vidgen, or /vanimate render, the bot shows what your idea was expanded into, with â
Render, âŠī¸ My text, and â Cancel buttons. Reply to that message with your own version and it renders that instead. Only the person who asked can decide./ask works as a reply: reply a bare /ask to any text message and that message becomes the question./musicgen <song idea> (/mg, /music): a full song with vocals from a one-line idea. Prefix a length like 120s: (up to 300s); raw: uses your text as the music description with no lyrics./ask answers come from an upgraded main AI that never refuses a question. local: still forces the backup AI./findimage <words> (/fi): search the trigger-picture library by word or tag and get the matching pictures back, each captioned with the word that fires it./help now just sends the link to the command guide instead of the wall of text./enhance: up-res a photo. It comes back much sharper and larger, delivered as a file so Telegram does not recompress it. Extra words steer it./memeit renders a short video with sound instead of a still image./ask got a much smarter answer model. local: forces the free backup AI (it never refuses), gpt: forces the main one./gifgen renders faster with richer motion./vidgen or /vanimate while your first is still rendering is refused instead of queued. (/musicgen and video /memeit share the same rule.)/vidgen <prompt> (/vg) and /vanimate (/va): short videos with sound. Voices, sound effects, and music are generated together with the picture, so describe what you want to hear too./mutetrigger <word> and /unmutetrigger <word>: silence a trigger picture in one chat only; all mutes everything there. /images marks muted entries with đ./help links it./help had silently stopped working because the text outgrew Telegram's message limit./gifgen <prompt> (/gg) and /animate: generate a ~5 second GIF from a prompt, or bring a photo to life. Prompts are expanded into motion beats first; raw: renders exactly what you typed.flux2: image model for /imagegen: understands lazy one-line prompts the way the cloud models do./ask answers straight by default now; meme: brings the joke delivery back per question./addimage <word> (/addpic) saves a photo or GIF under a word, and from then on that word posts it back. The bot tags what it sees in the picture; /images lists the library, /delimage removes one. Anyone can set them off./ask no longer attaches web pictures on its own; pics: asks for them./imagegen and /edit. removedThe mage: image model./ask <question>: ask anything. It searches the web when a question needs current information and cites its sources. web: forces a search, nosearch: skips it, pics: attaches real web pictures, meme: / serious: pick the delivery. Works on photos too: reply /ask <question> to one./merge [instruction]: combine two photos into one image (reply to the first, attach the second, or send both as an album)./joke, and /joke news for one about today's headlines./imagegen models: qwen: (best at legible text in the image), flash:, and dev: (FLUX.1 dev); plus real style transfer for qwen: with anime:, realism:, and pixel:./gif./imagegen <prompt> (/ig): image generation with krea:, flux:, zimage:, and the cloud models gemini: and gpt:. If a cloud model declines a prompt the bot re-renders it locally and says so. Style presets like deepfried: and epic:; a bare /imagegen lists them./edit <instruction> (/ie): change only what you ask in a photo you send or reply to./memeit: turn everything said in the group since the last /memeit into a meme./ascii <name> and /lenny emoticons./gif <search> and inline mode (@naertnetworksGIFBOT <search> in any chat).