First decide whether the guitar should disappear or be alone
How to remove guitar from a song is two different requests that share a search phrase. One request is a mix with the guitar turned down or gone, so a vocal, a horn, or your own part can occupy that space. The other request is the guitar by itself, so you can learn the part, sample a chord, or drop it into a new arrangement. The first wants the guitar subtracted from the song. The second wants the guitar kept and everything else subtracted. A single button labeled remove guitar does not tell you which file you will get.
Both requests fail in the same way if you only have a vocal remover. That tool estimates the singer and writes everything else to one instrumental. The rhythm guitar, the lead line, the piano comp, the bass, and the drums are still in that instrumental, at roughly the balance they had in the mix, minus whatever the model stole or left behind. You have removed the vocal. You have not removed the guitar. The stem splitter FAQ answers this directly: drums and bass are not exported on their own. If your notes say the guitar is the problem, a vocal/instrumental split will sound like success in the vocal file and like no change in the bed.
The case that does work today is the one where the vocal is the part you wanted out, and the guitar is supposed to stay. Then the instrumental is the right file. Use it as a practice bed or a remix reference. Do not expect a later gain move on that file to peel the guitar off the piano. They were never written to separate tracks.
A sequence that matches the file you actually have
Start from the earliest stereo file you are allowed to use. A session bounce or a lossless copy beats a re-encoded MP3. Trim to the section where the guitar matters: a riff, a chorus, a bar of the solo. A two-minute excerpt is enough to judge a method, and it is also the free duration cap on this site.
Name the output before you process anything. Write down one sentence: guitar gone, other parts stay, or guitar alone. If the sentence is guitar gone and you do not have the multitrack, you need a separator that offers a guitar stem, or you need to accept an EQ compromise. If the sentence is guitar alone, you need that same kind of stem, then you mute every stem that is not guitar. If the sentence is really the singer is in the way, stop and use a vocal remover. Mixing those sentences up is how people upload a song, download an instrumental, and think the guitar remover failed.
If you have the session, do not separate the master. Mute the guitar track, or solo it, and export. That is the only method here that does not guess. AI stem separation is the fallback when the session is gone.
If you do not have the session, listen to a short split from a tool that names the stem you need. For a guitar stem, the download list has to include guitar, not only vocals and instrumental. Four stems that stop at drums, bass, other, and vocals still leave most guitars inside other, beside the piano. Six names only help when guitar is one of the names. After it finishes, solo the guitar file and then mute it against the rest. Check a loud strum and a quiet tail. Bleed from the vocal or the snare means the stem is a guide, not a replacement track.
On this site, skip that guitar step, because the output list does not include it. The stem splitter will give you vocals and one instrumental. Use that only when the sentence you wrote was about the singer. The comparison of those two products is in AI stem splitter vs vocal remover.
Removing piano is the same kind of mistake
Piano sits in the middle of the arrangement even more often than guitar. The left hand shares space with bass. The right hand shares space with vocal and with guitar. A two-stem instrumental keeps all of that. People ask how to remove piano from a song for the same pair of reasons: they want the piano out so a vocal sits clearer, or they want the piano alone so they can learn the changes. Neither falls out of a vocal remover.
Treat piano as a second example of the guitar rule, not as a separate project. If a tool lists piano as a stem, judge it the way you would judge guitar: solo it, then mute it, and listen for the vocal that leaked into the sustain. If the tool does not list piano, no amount of renaming the instrumental will create it. This site does not list piano. It stays in the grouped instrumental with the guitar.
What EQ can do, and what it will ruin
Before stem models, the practical cut was EQ. Find the band where the guitar is loudest, dip it, and hope the vocal still sounds like a vocal. A narrow dip on a bright acoustic can take the pick attack down. It also takes the consonant edge of the singer and the top of a shaker if they share that range. A wider dip to hide a distorted guitar will hollow out the vocal and the snare. You can get a private rough mix that is less guitar-heavy. You do not get a guitar stem, and you do not get the original vocal back unharmed.
Phase tricks are worse for guitar than they were for a centered vocal. A guitar is often doubled, panned, or recorded in stereo. Inverting a channel may thin one side of a wide part and leave the other side, while also punching a hole in the bass. If the guitar was already centered and dry, you might reduce it and also reduce the vocal, which is the opposite of a useful karaoke bed. Listen in mono after any mid-side move. A trick that only works in headphones is not a removed guitar.
AI separation is the better guess when the instrument occupies the same frequencies as the vocal, because it uses more than one band. It is still a guess. The failure is bleed and a soft, unstable edge, not a polite EQ dip. Use EQ when you need a fast level change and can tolerate collateral damage. Use a named stem when you need the instrument as its own file. Use the original track when you have it.
Phase, quality, and the rights that do not change
A stem can sound thin even when the instrument is gone, because the model also removed room, string noise, or a doubled part you wanted to keep. Compare the result to the original at matched loudness. If the vocal in the guitar-muted mix sounds duller, the separator took vocal energy with the guitar. That is ordinary. It is a reason to keep the original mix muted on another track, not a reason to EQ the life back into a damaged vocal and call it finished.
Source quality shows up here as much as it does on a vocal split. A low-bitrate encode has already smeared the pick attack and the vocal consonants into the same noise. The model then has less to go on, and the guitar file, if you ever get one, carries more of the vocal. Start from the least-encoded copy. Do not transcode an MP3 to WAV and expect the missing attack to return.
Copyright is unchanged by the method. Removing a guitar, removing a piano, or removing a vocal does not license the song. A stem is still derived from the recording and the composition. A private practice take is not the same decision as a public upload, a sample release, or a paid sync. If you cannot clear it, do not publish it.